{"id":215593,"date":"2016-04-19T02:37:41","date_gmt":"2016-04-19T02:37:41","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-academic-program&#038;p=215593"},"modified":"2023-02-26T23:41:51","modified_gmt":"2023-02-27T07:41:51","slug":"fellowships-microsoft-research-asia","status":"publish","type":"msr-academic-program","link":"https:\/\/www.microsoft.com\/en-us\/research\/academic-program\/fellowships-microsoft-research-asia\/","title":{"rendered":"Fellowship at Microsoft Research Asia"},"content":{"rendered":"\n\n<div class=\"wp-block-buttons is-layout-flex wp-block-buttons-is-layout-flex\"><\/div>\n\n\n\n\n\n\n<h2 id=\"about-the-program\">About the program<\/h2>\n\n\n\n<p><strong>Application deadline<\/strong>: June 10, 2022, 11:59 PM Beijing Time<\/p>\n\n\n\n<p>The Microsoft Research Asia Fellowship Program aims to empower and encourage PhD students in the Asia-Pacific region to realize their potential in computer science-related research.<\/p>\n\n\n\n<p>The Microsoft Research Asia Fellowship Program identifies the next generation of research leaders through a unique program that offers a combination of mentorship, research, networking, and academic opportunities to promising young candidates. Since its inception in 1999, the program has attracted over one thousand PhD candidates from top universities in the Asia-Pacific region, and 452 outstanding PhD students have been awarded the Microsoft Research Asia Fellowship honor. Fellows have gone on\u202fto become prominent researchers and influential individuals in academia and industry.<\/p>\n\n\n\n<p>Fellows earn more than a simple scholarship; they enjoy\u202flong-term close engagement with Microsoft Research Asia, one of the leading computer science research labs in the world.\u202fEach winner has the opportunity to complete an internship, during which they participate in hands-on, advanced research at <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-asia\/\">Microsoft Research Asia<\/a> in Beijing, China.<\/p>\n\n\n\n<p>For&nbsp;questions about this program, please contact <a href=\"mailto:fellowRA@microsoft.com\">fellowra@microsoft.com<\/a>.<\/p>\n\n\n\n<h3 id=\"2022-award-schedule\">2022 Award schedule<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>June 10, 2022 \u2013 Application deadline&nbsp;<\/li>\n\n\n\n<li>Mid-August \u2013 Online interview (only for a few candidates who pass 1st round review)<\/li>\n\n\n\n<li>September 2022 \u2013 Fellows announced<\/li>\n<\/ul>\n\n\n\n<h3 id=\"provisions-of-the-award\">Provisions of the award<\/h3>\n\n\n\n<p>The Microsoft Research Asia Fellowship Program offers some of the industry\u2019s most competitive incentives and rewards, including:&nbsp;<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Cash awards of $10,000 USD, which can be used to support research initiatives and academic endeavors including, but not limited to: attending international research conferences\/events, and subsidizing paper and patent application fees.&nbsp;<\/li>\n\n\n\n<li>Opportunity for intensive and individualized mentorship programs with a dedicated senior researcher, to help facilitate the fellow\u2019s research until graduation.<\/li>\n\n\n\n<li>Opportunity to attend the Microsoft Research Asia Academic Conference.<\/li>\n\n\n\n<li>At the discretion of Microsoft Research Asia, advisors of fellows might be invited to participate in collaborative research projects and academic exchange events.<\/li>\n<\/ul>\n\n\n\n<h2 id=\"eligibility-criteria\">Eligibility criteria<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Applicants must attend a university in mainland China, Hong Kong, Japan, Korea, Singapore, or Taiwan, and do research work that relates to computing topics in which Microsoft Research is concerned with (click on Research Areas at the top of the page for a full list).<\/li>\n\n\n\n<li>According to the definition of the normative period of PhD program of the applicant\u2019s university and region, the applicant must graduate in 2024 or later. The applicants and their advisors would be asked for supporting documents if needed.<\/li>\n\n\n\n<li>Microsoft Research Asia Fellowship Committee owns the final interpretation of above clauses.<\/li>\n<\/ul>\n\n\n\n<h2 id=\"supplementary-information\">Supplementary information<\/h2>\n\n\n\n<p>Microsoft has always been committed to the mission of empowering every person and every organization on the planet to achieve more and has actively promoted the extensive application of computer technology in various fields and in public welfare undertakings. With the progress of data acquisition, data processing, and artificial intelligence in recent years, computer technology is playing an increasingly important role in the research of basic science. The current problems and challenges faced by human society also require the urgent intersection and integration of basic scientific research and computer science. Based on the current trends and demand, the 2022 MSRA Fellowship will continue to welcome applicants in the field of <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/en.wikipedia.org\/wiki\/Computational_science\" target=\"_blank\" rel=\"noopener\">Computational Science<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> in order to discover academic talents who are innovating computer theory and technology to promote the development of basic scientific research, and to support and encourage them to better carry out their potential research work.<\/p>\n\n\n\n<h3 id=\"related-computational-science-subject-areas\">Related Computational Science subject areas<\/h3>\n\n\n\n<p>Please refer to the Research Areas > Other Sciences menu in the top navigation of this site to learn more about these research areas:<\/p>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/collaboration\/studies-in-pandemic-preparedness\/\">Studies in Pandemic Preparedness<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/project\/stationb\/\">Station B<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/structural-and-computational-insights-into-the-sars-cov-2-omicron-rbd-ace2-interaction\/\">Structural and computational insights into the SARS-CoV-2 Omicron RBD-ACE2 interaction<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/improved-drug-target-interaction-prediction-with-intermolecular-graph-transformer\/\">Improved Drug-target Interaction Prediction with Intermolecular Graph Transformer<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/loss-of-spike-n370-glycosylation-as-an-important-evolutionary-event-for-the-enhanced-infectivity-of-sars-cov-2\/\">Loss of Spike N370 glycosylation as an important evolutionary event for the enhanced infectivity of SARS-CoV-2<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/exploring-the-regulatory-function-of-the-n-terminal-domain-of-sars-cov-2-spike-protein-through-molecular-dynamics-simulation-adv-theory-simul-10-2021\/\">Exploring the Regulatory Function of the N\u2010terminal Domain of SARS\u2010CoV\u20102 Spike Protein through Molecular Dynamics Simulation<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/samf-a-self-adaptive-protein-modeling-framework-2\/\">SAMF: a self-adaptive protein modeling framework<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/complementing-sequence-derived-features-with-structural-information-extracted-from-fragment-libraries-for-protein-structure-prediction\/\">Complementing sequence-derived features with structural information extracted from fragment libraries for protein structure prediction<\/a><\/li>\n<\/ul>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/learning-physics-informed-neural-networks-without-stacked-back-propagation\/\">Learning Physics-Informed Neural Networks without Stacked Back-propagation<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/message-passing-neural-pde-solvers\/\">Message Passing Neural PDE Solvers<\/a>&nbsp;<\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/lie-point-symmetry-data-augmentation-for-neural-pde-solvers\/\">Lie Point Symmetry Data Augmentation for Neural PDE Solvers<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/solving-quantum-many-body-problem-artificial-neural-networks\/\">Solving the Quantum Many-Body Problem with Artificial Neural Networks<\/a><\/li>\n<\/ul>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/deep-learning-for-prediction-of-the-air-quality-response-to-emission-changes\/\">Deep learning for prediction of the air quality response to emission changes<\/a>&nbsp;<\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/mimicking-atmospheric-photochemical-modeling-with-a-deep-neural-network\/\">Mimicking atmospheric photochemical modeling with a deep neural network<\/a>&nbsp;<\/li>\n\n\n\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/exploring-deep-learning-for-air-pollutant-emission-estimation\/\">Exploring deep learning for air pollutant emission estimation<\/a><\/li>\n<\/ul>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/project\/graphormer\/\">microsoft\/Graphormer<\/a> \u2013 a deep learning package that allows researchers and developers to train custom models for molecule modeling tasks. It aims to accelerate the research and application in AI for molecule science, such as material design, drug discovery, etc.<\/li>\n\n\n\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2110.06197\" target=\"_blank\" rel=\"noreferrer noopener\">Crystal Diffusion Variational Autoencoder for Periodic Material Generation<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/li>\n\n\n\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.science.org\/doi\/full\/10.1126\/sciadv.abf7435\" target=\"_blank\" rel=\"noreferrer noopener\">Accelerated discovery of 3D printing materials using data-driven multi-objective optimization<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/li>\n<\/ul>\n\n\n\n\n\n<h3 id=\"evaluation-criteria-of-computational-science-area-applicants\">Evaluation criteria of computational science area applicants<\/h3>\n\n\n\n<p>In the Computational Science area, The MSRA Fellowship program focuses on doctoral students who have promoted and expanded research work in the above fields through innovations in computer theory, algorithms, and systems and have made influential achievements. We also look at doctoral students who have made contributions to the development of computer technology by introducing the research methods, theories, and technologies of basic science into the computer science field.<\/p>\n\n\n\n\n\n<h2 id=\"how-to-apply\">How to apply<\/h2>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-1 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:66.66%\">\n<p>The 2022 Microsoft Research Asia Fellowship <strong>application period closes on Friday, June 10, 2022, at 11:59 PM Beijing Time<\/strong>.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>PhD students must apply directly for the scholarship.<\/li>\n\n\n\n<li>Applications must include:\n<ul class=\"wp-block-list\">\n<li>An application form<strong>&nbsp;<\/strong>(<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/04\/MSRA_Fellowship_Application_Form-2022.docx\" target=\"_blank\" rel=\"noreferrer noopener\">download template here<\/a>)<\/li>\n\n\n\n<li>Curriculum vitae<\/li>\n\n\n\n<li>Research Statement. This statement (of approximately 1,000 words) provides a summary of your research accomplishments and current work and highlight your most outstanding contribution to the academia. It must discuss the future direction and potential of your work, especially your research plan for the next few years of PhD study.<\/li>\n\n\n\n<li>At least two letters of reference from established researchers familiar with your research. Of these, one letter should come from your advisor. Scanned copies of the signed recommendation letters should be sent, by those recommending you, directly to <a href=\"mailto:fellowra@microsoft.com\">fellowra@microsoft.com<\/a>. <\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Applications must be submitted via the <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/cmtint.research.microsoft.com\/User\/Login?ReturnUrl=%2FMSRAFellowship2022\" target=\"_blank\" rel=\"noreferrer noopener\">online application tool<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> in any of the following formats: Word document, text-only file, PDF or ZIP. Email or hard copy applications will not be considered.<\/li>\n\n\n\n<li>Applications submitted to Microsoft will not be returned. Microsoft cannot assume responsibility for the confidentiality of information in submitted applications. Therefore, applications should not contain information that is confidential, restricted, or sensitive.<\/li>\n\n\n\n<li>Incomplete applications will not be considered.<\/li>\n\n\n\n<li>Due to the volume of submissions, Microsoft Research Asia cannot provide individual feedback on applications that do not receive fellowship awards.<\/li>\n<\/ul>\n<\/div>\n\n\n\n<div class=\"wp-block-column has-light-blue-background-color has-background is-layout-flow wp-block-column-is-layout-flow\" style=\"padding-top:20px;padding-right:20px;padding-bottom:20px;padding-left:20px;flex-basis:33.33%\">\n<h4 id=\"timeline\">Timeline<\/h4>\n\n\n\n<p><em>Program dates subject to change and will be updated here as needed.<\/em><\/p>\n\n\n\n<p><strong>April 20, 2022<\/strong>: Application period opens<br><strong>June 10, 2022<\/strong>: Application period closes<br><strong>Mid-August 2022<\/strong>: Online interviews<br><strong>September 2022<\/strong>: Selected applicants announced<\/p>\n<\/div>\n<\/div>\n\n\n\n\n\n<h2 id=\"frequently-asked-questions\">Frequently asked questions<\/h2>\n\n\n\n\n\n<p>The research statement should provide a summary of your research accomplishments and current work and highlight your most outstanding contribution to the academia. Importantly, it should clearly discuss the future direction and potential of your work, especially your research plan for the next few years of PhD study. It should be technical but remain intelligible to any reviewer of the track (we have four general tracks, which you can find in the online application tool). Because your research statement may be read by a reviewer with same track but outside of your subdiscipline, it is important to keep the \u201cbig picture\u201d in mind. A strong research statement presents a readable, compelling, and realistic research agenda with clear potential results. Research statements can be weakened by overly ambitious proposals, a lack of clear direction, and a lack of big-picture focus.<\/p>\n\n\n\n<p><strong>Some general advice:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>The goal of the research statement is to introduce yourself to a search committee and make them excited about your research. This committee will likely contain scientists both in and outside of your subdiscipline. The statement may be around 1,000 words, keeping in mind that <em>you want people to read it and buy in, <\/em>so please don\u2019t make it too long or too short<em>. <\/em><\/li>\n\n\n\n<li>Think of the overarching theme guiding your main research subject area. Write an essay that lays out:\n<ul class=\"wp-block-list\">\n<li>The main theme(s), why it is important, and what specific skills you will use to attack the problem.<\/li>\n\n\n\n<li>A couple of specific examples of problems you have already worked on with success \u2014&nbsp;to build credibility and give people an idea of what it is you do.<\/li>\n\n\n\n<li>A discussion of the future direction of your research. This section should build on the above and be exciting to people both in and outside your field. If you think that your research could lead to answers for big, exciting questions \u2014&nbsp;say so and reach for the stars.<\/li>\n\n\n\n<li>Tie it all off with a final paragraph that leaves the reader with a good overall impression of your research.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>There is no need to mention all of your projects; stick to those that align with your overarching theme. Introducing no more than three (3) research works is strongly recommended.<\/li>\n\n\n\n<li>Pay attention to jargon. You want reviewers to understand everything in your statement. Remember that the goal is to make the search committee excited about you \u2014&nbsp;and they won\u2019t be excited about something they can\u2019t understand.<\/li>\n\n\n\n<li>If you have something that sets you apart (e.g., a prestigious achievement in your field), you may want to include it.<\/li>\n\n\n\n<li>There are no excuses for spelling errors.<\/li>\n<\/ul>\n\n\n\n\n\n<p>Required application materials can be submitted directly online by the applicant. The online application tool is available at: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/cmtint.research.microsoft.com\/MSRAFellowship2022\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/cmtint.research.microsoft.com\/MSRAFellowship2022<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n\n\n<p>All application materials must be in English. All documents submitted must be in Word document, text-only file, PDF, or ZIP file format.<\/p>\n\n\n\n\n\n<p>Students must be doing research work that relates to computing topics in which Microsoft Research is concerned with (click on Research Areas at the top of the page for a full list).<\/p>\n\n\n\n<p><strong>Four major directions for reference:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Human machine intelligence <ul><li>Big data mining<\/li><li>Knowledge mining<\/li><li>Machine learning<\/li><li>Natural language processing<\/li><li>Social computing<\/li><li>Theory<\/li><\/ul>\n<ul class=\"wp-block-list\">\n<li>Web search and data mining<\/li>\n\n\n\n<li>Others <\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Perception, recognition, interaction\n<ul class=\"wp-block-list\">\n<li>Graphics<\/li>\n\n\n\n<li>Multimedia<\/li>\n\n\n\n<li>Robotics<\/li>\n\n\n\n<li>Speech<\/li>\n\n\n\n<li>Vision<\/li>\n\n\n\n<li>Others <\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Hardware and software systems\n<ul class=\"wp-block-list\">\n<li>Applied algorithms<\/li>\n\n\n\n<li>Cloud computing<\/li>\n\n\n\n<li>Devices<\/li>\n\n\n\n<li>Mobile sensing<\/li>\n\n\n\n<li>Networking<\/li>\n\n\n\n<li>Software analytics<\/li>\n\n\n\n<li>Systems<\/li>\n\n\n\n<li>Others <\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Computational science\n<ul class=\"wp-block-list\">\n<li>Computational biology and life science<\/li>\n\n\n\n<li>Computational physics<\/li>\n\n\n\n<li>Computational sustainability and environmental science<\/li>\n\n\n\n<li>Computational chemistry and material science<\/li>\n\n\n\n<li>Other Computational Science Fields<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\n\n<p>First-year students are eligible to apply if they have already enrolled in a PhD program by the time of the nomination.<\/p>\n\n\n\n\n\n<p>It depends on the expected graduation date of the normative PhD study you are enrolled in. If the student will graduate in 2024 or later, then they are eligible.<\/p>\n\n\n\n\n\n<p>Selected applicants will receive notification of their acceptance status by early October 2022. Due to the volume of submissions, Microsoft Research Asia cannot provide individual feedback on applications that do not receive fellowship awards.<\/p>\n\n\n\n\n\n<p>The funds are given as an unrestricted gift. Fellowship recipients are not subject to intellectual property restrictions unless they complete an internship at Microsoft Research Asia. If that is the case, they are subject to the same intellectual property restrictions as any other Microsoft Research Asia intern.<\/p>\n\n\n\n\n\n<p>Microsoft Research Asia Fellowship only accepts applicants with universities in mainland China, Hong Kong, Japan, Korea, Singapore, and Taiwan.<\/p>\n\n\n\n\n\n<p>Please send email directly to <a href=\"mailto:fellowRA@microsoft.com\">fellowRA@microsoft.com<\/a> for other questions<\/p>\n\n\n\n\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-hanzhi-wang.jpg\" alt=\"a woman smiling and posing for the camera\" class=\"wp-image-881820\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-hanzhi-wang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-hanzhi-wang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"hanzhi-wang-opens-in-new-tab\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/wanghzccls.github.io\/\">Hanzhi Wang<\/a><\/h3>\n\n\n\n<p>Renmin University of China<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/weizhewei.com\/\">Zhewei Wei<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Large-Scale Graph Analysis and Learning<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Driven by the exponential blowup in data volumes, efficient algorithms are now in higher demand more than ever before. The massive data volume also challenges the classical notion of efficient algorithms.&nbsp; Characterizing efficient algorithms in terms of exponential\/polynomial time complexity may no longer be sufficient for solving today&#8217;s problems. In light of the pressing needs for the algorithms with high scalability, the core of my research primarily focuses on the development of provably-good scalable algorithms for graph data. In particular, I am interested in designing nearly linear or sub-linear time algorithms to efficiently compute results for large-scale graph analysis and learning problems.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-haotian-li.jpg\" alt=\"a man smiling for the camera\" class=\"wp-image-881823\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-haotian-li.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-haotian-li-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"haotian-li-opens-in-new-tab\"><a href=\"https:\/\/haotian-li.com\/\">Haotian Li<\/a><\/h3>\n\n\n\n<p>The Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor:&nbsp;<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/huamin.org\/\">Huamin Qu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Data Visualization, Visual Analytics, and Human-Computer Interaction<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Recent years have witnessed an explosion of data, both in type and quantity. However, humans\u2019 ability to analyze data does not match such fast growth. To address the challenge, intelligent visual analysis tools have gained growing interest since they enable accurate and rapid data sensemaking by leveraging humans\u2019 high-bandwidth vision systems. My research aims to promote human-in-the-loop intelligent visual analysis with techniques from visualization, human-computer interaction, and machine learning. To achieve the goal, I devoted myself to understanding real-world practices of visualizations for data analysis and leveraging the findings to develop intelligent visual analysis tools. By combining the efforts in the two directions, I hope to contribute to an ecosystem where humans can perform effortless and adaptive visual data analysis with the support of machine intelligence. With the ecosystem, everyone can analyze data visually without barriers.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-heecheol-kim.jpg\" alt=\"pic\" class=\"wp-image-881826\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-heecheol-kim.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-heecheol-kim-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"heecheol-kim-opens-in-new-tab\"><a href=\"https:\/\/sites.google.com\/view\/heecheol-kim\/home\">Heecheol Kim<\/a><\/h3>\n\n\n\n<p>The University of Tokyo<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.isi.imi.i.u-tokyo.ac.jp\/\">Yasuo Kuniyoshi<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>&nbsp;<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Robotics<\/p>\n\n\n\n<p><strong>Long-term research goal: <\/strong>Recently, robotics has been gaining attention as a solution for the aging society that many countries are facing. Even though there is a high demand for automation of tedious work that now human labor is doing, robots&#8217; skills are premature and face difficulty for simple tasks. My research goal is to construct a unified robot learning architecture that can imitate human skills. To achieve this, I have focused on attention-based deep imitation learning inspired by physiological studies of humans. My research has achieved many dexterous robot tasks such as banana peeling, needle threading, and knot tying. My long-term research will focus on the unsupervised discovery of robot behavior\u2019s segmentation and structure that can be used for data-efficient robot training and adaptation. As a result of this study, I expect the robot to free humans from tedious, repetitive, and dangerous labor and make humans focus on more creative jobs.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-kun-zhou.jpg\" alt=\"a young man wearing a suit and tie smiling at the camera\" class=\"wp-image-881829\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-kun-zhou.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-kun-zhou-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"kun-zhou\">Kun Zhou<\/h3>\n\n\n\n<p>Renmin University of China<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/gsai.ruc.edu.cn\/addons\/teacher\/index\/info.html?user_id=30&ruccode=20130047&ln=en\">Ji-Rong Wen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/playbigdata.ruc.edu.cn\/batmanfly\/\">Wayne Xin Zhao<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>&nbsp;<\/p>\n\n\n\n<p><strong>Research interests:&nbsp;<\/strong>Natural Language Process, Information Retrieval<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;My research mainly focuses on learning effective representations for sequence data (e.g., textual sentences and user behaviors), which can be widely used on a variety of NLP and IR tasks. Despite the remarkable progress in recent years, there are two challenging problems to be resolved: how to learn transferable, scalable and robust representations from large-scale unsupervised sequence data, and how to capture the latent high-level knowledge and logic within the available sequence data. My long-term research goal is to address the two problems, and learn general, knowledgeable and practical sequence representations for various real-world applications.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-linfeng-zhang.jpg\" alt=\"pic\" class=\"wp-image-881832\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-linfeng-zhang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-linfeng-zhang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"linfeng-zhang\">Linfeng Zhang<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: Kaisheng Ma<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Knowledge Distillation, Efficient Machine Learning<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;My research interest is knowledge distillation, a deep learning technique to compress the neural networks, making them small enough to be deployed on the edge devices. Current deep neural networks usually have enormous parameters and computation which has hindered their usage on edge devices for real-world applications. I hope my research can bridge the gap between the large ML models and small edge devices, and also bridge the gap between our life and an intelligent future.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-ruyi-ji.jpg\" alt=\"pic\" class=\"wp-image-881835\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-ruyi-ji.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-ruyi-ji-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"ruyi-ji\">Ruyi Ji<\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisors: Zhenjiang Hu, Yingfei Xiong&nbsp;<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Program Synthesis<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Algorithms are crucial for improving the efficiency of programs, and studying algorithms has become a must thing for people related to programming and computer science. Despite their importance, algorithms are well-known to be complex. On the one hand, learning algorithms is a big challenge for most programmers. On the other hand, optimizing via algorithms is error-prone in practice. Such optimization can greatly increase the complexity, break the modularity, and thus bring risks of flaws. My research goal is to automatically synthesize efficient algorithms and thus provide a safe and mechanical method for performing algorithmic-level optimizations.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-wenjing-wang.jpg\" alt=\"pic\" class=\"wp-image-881838\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-wenjing-wang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-wenjing-wang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"wenjing-wang-opens-in-new-tab\"><a href=\"https:\/\/daooshee.github.io\/website\/\">Wenjing Wang<\/a><\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.icst.pku.edu.cn\/struct\/people\/liujiaying.html\">Jiaying Liu&nbsp;<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Computer Vision, Image Restoration<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Image restoration is to recover an image from a corrupted version. Traditionally, it is designed to meet users&#8217; demands on subjective visual quality, i.e., human vision. As intelligent software is replacing humans in works of image analysis, more and more images are used for downstream machine learning tasks, i.e., machine vision. However, most of existing image restoration methods neglect machine vision, which poses threats to applications using restored images for further analysis. My long-term research goal is to bridge the gap between human vision and machine vision. I aim to deepen the understanding of low-level and high-level computer vision, and pave a new way for the research of image restoration.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-xuanhe-zhou.jpg\" alt=\"a man wearing a suit and tie\" class=\"wp-image-881841\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-xuanhe-zhou.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-xuanhe-zhou-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"xuanhe-zhou-opens-in-new-tab\"><a href=\"https:\/\/db.zhouxh.store\">Xuanhe Zhou<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/dbgroup.cs.tsinghua.edu.cn\/ligl\/publications.html\">Guoliang Li<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, Jianhua Feng<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Autonomous Database Systems<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Database optimization techniques have been studied for over 50 years, which is like the \u201cmoat\u201d of database kernels, i.e., reducing the cost of ownership, optimizing the performance, and even improving the system robustness. However, most database products only adopt heuristic tools, which are manually crafted and have limited optimization capability. By integrating ML techniques, databases can automatically decide the logical and physical designs based on workload\/system characters, and potentially turns from \u201cdriver-assisted\u201d to \u201cself-driving\u201d. Thus, my long-term goal is to build a practical \u201cself-driving\u201d database system, which is equipped with three critical characters (i.e., proactive monitoring, configuration tuning, and learned optimizer), and can efficiently adapt to various SLA requirements.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yuji-roh.jpg\" alt=\"pic\" class=\"wp-image-881844\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yuji-roh.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yuji-roh-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"yuji-roh-opens-in-new-tab\"><a href=\"https:\/\/www.yujiroh.com\/\">Yuji Roh<\/a><\/h3>\n\n\n\n<p><strong>KAIST <\/strong><strong><\/strong><\/p>\n\n\n\n<p><strong>Supervisors: <\/strong><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/stevenwhang.com\/\"><strong>Steven Euijong<\/strong><span class=\"sr-only\"> (opens in new tab)<\/span><\/a> <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/stevenwhang.com\/\"><strong>Whang<\/strong><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Trustworthy AI, Model Fairness, Human-Centered AI<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Trustworthy AI is becoming indispensable for modern machine learning applications. Among various trustworthiness aspects, I am especially interested in building fair AI frameworks without demographic disparities. As my long-term research goal, I aim to bridge the gap between fair AI technology and the real world by building human-centered fair AI systems. Most of the existing fairness algorithms assume mathematical definitions of fairness, which do not fully capture various social contexts. To address this problem, I aim to build a human-centered fair training system that better reflects humans needs of fairness into any application. I envision that this research will take us a step closer to positively impacting society.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yulin-wang.jpg\" alt=\"pic\" class=\"wp-image-881847\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yulin-wang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yulin-wang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"yulin-wang-opens-in-new-tab\"><a href=\"https:\/\/www.rainforest-wang.cool\/\">Yulin Wang<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.cae.cn\/cae\/html\/main\/colys\/96944507.html\">Cheng Wu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>; <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.gaohuang.net\/\">Gao Huang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Dynamic Neural Networks; Efficient Deep Learning; Computer Vision<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Human brains are far more efficient than current deep learning models. The brains can learn from a small amount of multi-modal data (e.g., vision, language, and audio) with minimal power consumption. The inference process of brains is quite cheap such that it can be supported with a little chemical energy. These advantages may come from the sparsity of human brains: only a few specialized neurons are activated and trained for each specific task. The long-term goal of my research is to develop the sparsely-activated deep networks that work like human brains. My ambition is to close the gap from AI to human intelligence in training\/inference efficiency by exploring this direction.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yuqi-li.jpg\" alt=\"pic\" class=\"wp-image-881850\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yuqi-li.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-yuqi-li-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"yuqi-li-opens-in-new-tab\"><a href=\"https:\/\/www.researchgate.net\/profile\/Yuqi-Li-7\">Yuqi Li<\/a><\/h3>\n\n\n\n<p>University&nbsp;of&nbsp;Chinese&nbsp;Academy&nbsp;of&nbsp;Sciences<\/p>\n\n\n\n<p>Supervisors: Yong-Sheng Hu<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Energy storage, Machine Learning<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Rechargeable batteries are crucial in many applications ranging from portable electronics and medical devices, to renewable energy integration in power grids and electric vehicles. It is important to predict the properties or improve performance of these batteries via in situ monitoring methods without damage. Thus, an intelligent battery management system is necessary. However, the chemical\/physical processes that take place inside a battery cell during operation are very complex. The overall goal of my research is to combine the electrochemical mechanisms and data-driven model to explore the underlying common laws of rechargeable batteries and further achieve dynamic battery life prediction and optimization.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<p><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-ze-liu.jpg\" alt=\"pic\" class=\"wp-image-881853\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-ze-liu.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2022\/09\/2022-fellows-ze-liu-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"ze-liu-opens-in-new-tab\"><a href=\"https:\/\/zeliu98.github.io\/\">Ze Liu<\/a><\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors:&nbsp;Baining Guo, Yong Wang<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Computer Vision, Foundation Model<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;My research has focused on developing basic neural architectures for computer vision, particularly the general-purpose visual backbone which extracts image\/video features and is applicable to various computer vision tasks. Visual backbone is the foundation for various visual problems and the improvements of backbone architecture will benefit almost all vision tasks. In the future, I will further explore how to build a more generic vision architecture that can handle almost all vision problems with various types of visual signals.<\/p>\n\n\n\n\n\n\n\n<p>Over one hundred fifty distinguished PhD candidates from 50 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2021 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 11 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.<\/p>\n\n\n\n<p>Microsoft Research Asia recognizes the following fellows, who represent the best and the brightest PhD candidates in the Asia-Pacific region.<\/p>\n\n\n\n<div class=\"wp-block-group is-layout-flow wp-block-group-is-layout-flow\">\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-dianyu-chen.jpg\" alt=\"Dianyu Chen\" class=\"wp-image-786601\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-dianyu-chen.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-dianyu-chen-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"dianyu-chen\"><strong>Dianyu Chen<\/strong><\/h3>\n\n\n\n<p>Westlake University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/sls.westlake.edu.cn\/en\/Faculty\/201912\/t20191202_2407.shtml\">Heping Xu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Computational Biology, Humoral Immunity<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;The humoral immune response is a type of adaptive immune response that enables the human body to defend itself in a targeted way, and is the basis for the vaccine to work. Despite considerable progress, there remain many knowledge gaps in the understanding of the cellular and molecular mechanisms underlying humoral immunity, which hinders the production of effective vaccines against many deadly viruses. The overall goal of my research is to leverage computational biology approaches to deepen our understanding about humoral immunity, thus providing new prescriptions for the development of urgently needed vaccines. Moreover, such knowledge can also contribute to the development of therapies for autoimmune disease.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-kaizhang-kang.jpg\" alt=\"Kaizhang Kang\" class=\"wp-image-786607\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-kaizhang-kang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-kaizhang-kang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"kaizhang-kang-opens-in-new-tab\"><a href=\"http:\/\/cocoakang.cn\/\"><strong>Kaizhang Kang<\/strong><\/a><\/h3>\n\n\n\n<p>Zhejiang University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/hongzhiwu.com\">Hongzhi Wu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Appearance Acquisition, 3D scanning and Machine Learning<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;My long-term research goal is to answer this question: How to digitize arbitrary real objects with challenging appearance properties and complex geometry in an efficient and high-quality manner? In past years, my research reformulated the 3D modeling problem and proposed a new methodology to jointly optimize both acquisition & reconstruction procedures. By this, we proposed a differentiable acquisition framework as well as a series of high-performance setups. Although we can efficiently digitize most daily objects with high quality now, several properties remain hard to model, such as transparency. I am going to tackle these challenging problems in the future. In addition, enhancing medical imaging techniques by our novel acquisition framework is also an interesting topic to me.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-libo-qin.jpg\" alt=\"Libo Qin\" class=\"wp-image-786610\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-libo-qin.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-libo-qin-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"libo-qin-opens-in-new-tab\"><a href=\"http:\/\/ir.hit.edu.cn\/~lbqin\/\"><strong>Libo Qin<\/strong><\/a><\/h3>\n\n\n\n<p>Harbin Institute of Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ir.hit.edu.cn\/~car\/\">Wanxiang Che <span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests<\/strong>: Natural Language Processing and Task-oriented Dialogue System<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;My research mainly focuses on natural language processing and task-oriented dialogue system (ToDs), which help users to achieve specific goals. With the burst of deep neural networks and the evolution of pre-trained language models, the research of ToDs has obtained significant breakthroughs. However, the existing models heavily rely on large training data, which is only available for a certain number of task domains or languages, lacking scalability, practicality, and robustness. Inspired by the great success of transfer learning, my long-term research goal is to explore transfer learning for ToDs, aiming to build a scalable, practical and robust task-oriented dialogue system with limited data.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qing-wang.jpg\" alt=\"Qing Wang\" class=\"wp-image-786613\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qing-wang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qing-wang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"qing-wang-opens-in-new-tab\"><a href=\"http:\/\/storage.cs.tsinghua.edu.cn\/~wq\/\"><strong>Qing Wang<\/strong><\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: Jiwu Shu, Youyou Lu<\/p>\n\n\n\n<p><strong>Research interests<\/strong>: Memory\/Storage Systems<\/p>\n\n\n\n<p><strong>Long-term research goal<\/strong>: Shared memory systems underpin modern applications using a unified memory abstraction. My research rethinks shared memory systems, which is motivated by three hardware trends in datacenters, including new memory (e.g., persistent memory), new network (e.g., RDMA and CXL), and new compute (e.g., programmable switch ASIC). I have built several efficient shared memory systems at different system scales. My long-term research goal is abstracting the entire datacenter as a giant shared-memory machine, which can store, retrieve, and process data with high throughput, low latency, and good programmability.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-aoyu-wu.jpg\" alt=\"Aoyu Wu\" class=\"wp-image-786592\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-aoyu-wu.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-aoyu-wu-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"aoyu-wu-opens-in-new-tab\"><a href=\"https:\/\/wowjyu.github.io\/\"><strong>Aoyu Wu<\/strong><\/a><\/h3>\n\n\n\n<p>Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/huamin.org\/\">Huamin Qu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests<\/strong>: Data Science and Visualization, Human-Computer Interaction<\/p>\n\n\n\n<p><strong>Long-term research goal<\/strong>: Data visualizations often serve as the main entry point for the public to access data. My goal is to make visualizations a \u201cfirst-class citizen\u201d on the web and more accessible to both humans and computers. Specifically, I hope to build tools that automatically recommend models and visualizations for data analysis to advance the democratization of data science. Furthermore, unlike humans, machines have no direct access to the information inside visualizations. Massive numerical information and knowledge remain locked inside visualizations. I plan to create methods for interpreting and understanding the implicit meanings of visualizations at the Internet scale. By integrating those two perspectives, I aim to contribute to a new online knowledge ecology \u2013 both analyzing web visualizations to distill knowledge and assisting the public in producing new visualizations to communicate information. This ecology will increase the breadth of the semantic web and make \u201cbig data\u201d more accessible and efficient.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qitian-wu.png\" alt=\"Qitian Wu\" class=\"wp-image-786616\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qitian-wu.png 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qitian-wu-146x180.png 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"qitian-wu\"><strong>Qitian Wu<\/strong><\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor: Junchi Yan<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Machine Learning, Data Mining<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Modern machine learning systems need to interact with new unseen instances from a dynamic open world when adapted from training data to test observations or transferring from simulated environments to real situations. This requires the model\u2019s capability for in-the-wild generalization and extrapolation to new distributions. My research will focus on developing the next generation of representation learning approach that inherits the adaptability of human beings and can achieve inductive reasoning from familiar concepts to new ones. Here are two challenging problems to be resolved: how to define the relations among concepts and how to infer the latent structures among them. Such a new paradigm could benefit various tasks that need to obtain instance-specific embedding in broad AI areas and also shorten the gap between machines and human intelligence.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-hiromu-yakura.jpg\" alt=\"Hiromu Yakura\" class=\"wp-image-786604\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-hiromu-yakura.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-hiromu-yakura-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"hiromu-yakura-opens-in-new-tab\"><a href=\"https:\/\/yumetaro.info\/\"><strong>Hiromu Yakura<\/strong><\/a><\/h3>\n\n\n\n<p>University of Tsukuba<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/staff.aist.go.jp\/m.goto\/\">Masataka Goto<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Human-Computer Interaction, Machine Learning<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;As a researcher in HCI, my research focus is on expanding the application area of computers, especially, machine learning techniques. Given that we cannot deny the possibility of the mistakes of machine learning models, we need to carefully design the interaction between the models and humans to realize practical applications. In this context, I believe that understanding the characteristics of both machine learning models and humans is crucial. This is why I have associated my research with other interdisciplinary fields such as behavioral economics and cultural studies. I hope that I can reach a new perspective of understanding humans through exploring practical applications of machine learning in this human-centric manner.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-da-yu.jpg\" alt=\"Da Yu\" class=\"wp-image-786598\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-da-yu.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-da-yu-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"da-yu\"><strong>Da Yu<\/strong><\/h3>\n\n\n\n<p>Sun Yat-sen University<\/p>\n\n\n\n<p>Supervisors: Jian Yin, Tie-Yan Liu<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Privacy-preserving Learning, Trustworthy Deep Learning<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;My research aims to make deep learning safer to use. Despite its remarkable success, deep learning is prone to various of attacks, e.g., privacy attacks that steal user data and adversarial attacks that manipulate model outputs. Such attacks arouse concerns about deep learning and hinder its applications in real world. My future research goal is to build reliable algorithms to earn the trust of the participators from different aspects of deep learning.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wangyou-zhang.jpg\" alt=\"Wangyou Zhang\" class=\"wp-image-786622\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wangyou-zhang.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wangyou-zhang-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"wangyou-zhang-opens-in-new-tab\"><a href=\"https:\/\/x-lance.sjtu.edu.cn\/members\/wangyou-zhang\"><strong>Wangyou Zhang<\/strong><\/a><\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor: Yanmin Qian<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Speech Signal Processing, Robust Speech Recognition<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong>&nbsp;Speech is one of the most important ways in our daily communication. And \u201cone of our most important faculties is our ability to listen to, and follow, one speaker in the presence of others\u201d. However, the mechanism behind this ability and how machines can mimic human\u2019s ability in the complex real-world scenarios are still not well studied. This is known as the cocktail party problem, one of the most challenging problems in speech processing over the past 60 years. The advances in deep learning have greatly pushed forward the study in speech processing in the past decade. However, the cocktail party problem is still very difficult and remains unsolved. My long-term research goal is to address the multi-speaker speech modeling and processing in realistic scenarios and ultimately tackle the cocktail party problem.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wanjun-zhong.jpg\" alt=\"a young man wearing a suit and tie\" class=\"wp-image-786625\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wanjun-zhong.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wanjun-zhong-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"wanjun-zhong\"><strong>Wanjun Zhong<\/strong><\/h3>\n\n\n\n<p>Sun Yat-Sen University<\/p>\n\n\n\n<p>Supervisors: Ming Zhou, Jian Yin, Jiahai Wang<\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Machine Reasoning, Natural Language Processing<\/p>\n\n\n\n<p><strong>Long-term research goal<\/strong>: Current neural network models learn from large-scale data, and its working mechanism is a black box. Therefore, they highly depend on large-scale annotated data, and are unreliable because the prediction is hard to be explained. Besides, it has limited use of human knowledge. To alleviate these problems, my research goal focuses on building reliable, data-efficient, controllable AI systems utilizing world knowledge. Finally, I hope the machines can better assist human to solve difficult reasoning tasks with world knowledge, like debates, reliable medical diagnosis, or solving mathematical problems.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\" \/>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-tiankuang-zhou.jpg\" alt=\"Tiankuang Zhou\" class=\"wp-image-786619\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-tiankuang-zhou.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-tiankuang-zhou-146x180.jpg 146w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"tiankuang-zhou-opens-in-new-tab\"><a href=\"https:\/\/rickyim.github.io\/\"><strong>Tian<\/strong><strong>k<\/strong><strong>uang Zhou<\/strong><\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.luvision.net\/\">Lu Fang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/media.au.tsinghua.edu.cn\/\">Qionghai Dai<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong>&nbsp;Optical Intelligent Computing, Machine Learning<\/p>\n\n\n\n<p><strong>Long-term research goal<\/strong>: As the size of electronic transistor is approaching to its physical limit, electronic computer faces challenges in continuously improving the computing speed and energy efficiency. Modern computer science problems, however, requires drastically-increasing amount of compute. Endowed with high-speed and low-loss propagation, light is regarded as a promising candidate to remedy the conflict. My research focuses on employing photonics and optics to build novel computing architectures to solve computer science problems. My long-term goal is three-fold. Firstly, I want to build photonic computing machines that have the advantage to accelerate the inference process of artificial intelligence (AI). Secondly, since training of AI is time- and energy- consuming, I plan to construct efficient photonic systems for the training of AI. Thirdly, the computing process based on photons is fundamentally distinct from electrons. Because of this, I am striving for a photonic computing paradigm which can help us tackle the long-standing hard problems.<\/p>\n<\/div>\n\n\n\n\n\n\n\n<p>Over one hundred distinguished PhD candidates from 36 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2020 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 12 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.<\/p>\n\n\n\n<p>Microsoft Research Asia recognizes the following fellows, who represent the best and the brightest PhD candidates in the Asia-Pacific region.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghyun-Woo_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Sanghyun Woo\" class=\"wp-image-708853\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghyun-Woo_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghyun-Woo_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"sanghyun-woo-opens-in-new-tab\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/sites.google.com\/view\/sanghyunwoo\/\" target=\"_blank\" rel=\"noopener\">Sanghyun Woo<\/a><\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: In So Kweon<\/p>\n\n\n\n<p><b>Research interests:<\/b> Computer Vision, Machine Learning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Artificial Intelligence (AI) is on its way to ubiquity. There has been a flurry of advances in AI technologies, yet the current AI systems are still too narrow and passive than human-level intelligence. My research aims to build an AI agent that can think and learn like humans. To do so, I focus on challenging issues in the current learning system both in the \u2018model\u2019 and \u2018data\u2019 perspective. Specifically, there are two principal directions I have explored. First, apart from the current model design paradigm, I attempted to incorporate the attention mechanism, a central component of the human vision system, into the model. Second, to reduce expensive and error-prone human supervision, I tried to exploit self-supervision or simulation and scale the training data up to a billion-scale. Based on my accumulated research experiences in both aspects, the future directions aim to unify two distinct perspectives and present the general learning framework that can incubate various future AI technologies.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Hao-Shu-Fang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Hao-Shu Fang\" class=\"wp-image-708874\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Hao-Shu-Fang_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Hao-Shu-Fang_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"hao-shu-fang-opens-in-new-tab\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/fang-haoshu.github.io\/\" target=\"_blank\" rel=\"noopener\">Hao-Shu Fang<\/a><\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/mvig.sjtu.edu.cn\/\" target=\"_blank\" rel=\"noopener\">Cewu Lu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Human Understanding, Robotics Grasping\/Manipulation<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Imagine a world where robots can learn to cook a perfect steak by observing your demonstration. My life dream is to build robots that are smart enough to learn many of the things humans do: assembly, cooking or cleaning the table. And I envision robots to quickly learn manipulation skills from human demonstration. This requires a breakthrough in several fields of AI, including human understanding and robotics manipulation. My research goal focuses on answering the following questions: <em>How can machines acquire visual, geometric and physical attributes of humans and objects? How to obtain an appropriate representation of human\u2019s demonstration? How to jointly model visual and force perception for robot manipulation? How to define an objective function for a robot?<\/em> These interesting and challenging problems motivate me to go further in my research.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Daya-Guo_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Daya Guo\" class=\"wp-image-708868\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Daya-Guo_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Daya-Guo_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"daya-guo-opens-in-new-tab\"><a href=\"https:\/\/guoday.github.io\/\" target=\"_blank\" rel=\"noopener\">Daya Guo<\/a><\/h3>\n\n\n\n<p>Sun Yat-sen University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/sdcs.sysu.edu.cn\/content\/2505\" target=\"_blank\" rel=\"noopener\">Jian Yin<\/a>,\u202f<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/mingzhou\/\">Ming Zhou<\/a> (Microsoft)<\/p>\n\n\n\n<p><b>Research interests:<\/b> Natural Language Processing and Code Intelligence<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research mainly focuses on natural language processing and code intelligence to enable computers to intelligently process, understand and generate both natural language and programming language. With the growing population of developers, leveraging AI technology to increase the productivity of developers becomes more and more important in both communities of software engineering and artificial intelligence. Inspired by the great success of large pre-trained models like BERT and GPT in natural language processing, my long-term research goal is to develop powerful pre-trained models to support various code-related tasks, e.g. code search and code completion, to improve the productivity of the development process.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Ling-Pan_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Ling Pan\" class=\"wp-image-708880\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Ling-Pan_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Ling-Pan_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"ling-pan-opens-in-new-tab\"><a href=\"https:\/\/ling-pan.github.io\/\" target=\"_blank\" rel=\"noopener\">Ling Pan<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/people.iiis.tsinghua.edu.cn\/~huang\/\" target=\"_blank\" rel=\"noopener\">Longbo Huang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Reinforcement Learning, Computational Sustainability<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Recent years have witnessed great success of reinforcement learning (RL) with deep feature representations in many challenging tasks, including computer games, robotics, etc. However, it still faces several major challenges, especially when combined with deep neural networks. My research focuses on developing robust, efficient, and practical RL algorithms, and targets three core questions. Firstly, how can we ensure a convergent and robust learning behavior of an RL agent. Secondly, how can we tackle the high sample complexity problem of RL algorithms. Thirdly, how to successfully apply RL algorithms in important practical applications, e.g., computational sustainability problems.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghoon-Kang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Sanghoon Kang\" class=\"wp-image-708850\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghoon-Kang_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghoon-Kang_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"sanghoon-kang\">Sanghoon Kang<\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: Hoi-Jun Yoo<\/p>\n\n\n\n<p><b>Research interests:<\/b> On-Device AI, SW\/HW Co-Design<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Despite the mind-blowing abilities of AI algorithms, it is still difficult to deploy AI applications to people\u2019s everyday devices. The inevitable physical limitation of data gathering prevents AI models from performing at its best in different environments. On-device training, which can solve the adaptability issue, is strongly challenged by the limited hardware performance and battery constraints of edge devices. My research goal is to develop software & hardware approaches for energy-efficient AI computing, to enable edge devices with on-device training capabilities.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Juheon-Yi_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Juheon Yi\" class=\"wp-image-708877\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Juheon-Yi_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Juheon-Yi_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"juheon-yi-opens-in-new-tab\"><a href=\"https:\/\/juheonyi.github.io\/\" target=\"_blank\" rel=\"noopener\">Juheon Yi<\/a><\/h3>\n\n\n\n<p>Seoul National University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/youngkilee.blogspot.com\/\" target=\"_blank\" rel=\"noopener\">Youngki Lee<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Mixed Reality (MR), Mobile\/Embedded Deep Learning Systems<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> The goal of my research is to design futuristic MR apps, comprehensively analyze their workloads, and build novel enabling mobile systems to support them. Despite their vast potential, truly immersive MR apps are yet to be realized. The core challenge lies in the unique workload of seamlessly combining the virtual contents over the physical world on resource-constrained mobile\/wearable devices. So far, I have been working on projects to develop core mobile\/embedded deep learning systems for fully-immersive MR apps. To realize my vision, I plan to cover more diverse aspects of MR by conducting interdisciplinary research at the intersection of mobile computing, deep learning, multimedia, and HCI.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Mo-Zou_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Mo Zou\" class=\"wp-image-708883\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Mo-Zou_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Mo-Zou_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"mo-zou-opens-in-new-tab\"><a href=\"https:\/\/ipads.se.sjtu.edu.cn\/pub\/members\/mo_zou\" target=\"_blank\" rel=\"noopener\">Mo Zou<\/a><\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor:\u202fHaibo Chen<\/p>\n\n\n\n<p><b>Research interests:<\/b> Formal Verification, Operating System<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> System software such as operating systems is the backbone of computer systems. A single bug in operating systems could endanger the whole system. Formal verification can provide the highest level of assurance to system software. But applying formal verification to system software, especially concurrent ones, faces many challenges, from both theory and practice. My research tries to solve these challenges. My research goal is to make critical software systems truly reliable and secure through formal verification.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Yangbangyan-Jiang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Yangbangyan Jiang\" class=\"wp-image-708862\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Yangbangyan-Jiang_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Yangbangyan-Jiang_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"yangbangyan-jiang-opens-in-new-tab\"><a href=\"https:\/\/jiangyangby.github.io\/\" target=\"_blank\" rel=\"noopener\">Yangbangyan Jiang<\/a><\/h3>\n\n\n\n<p>University of Chinese Academy of Sciences<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/qmhuang-ucas.github.io\/\" target=\"_blank\" rel=\"noopener\">Qingming Huang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Deep Learning, Clustering<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Clustering provides us a way to release human efforts in annotation. Nowadays, we have witnessed the tremendous progress of clustering. However, the vast majority of existing methods only deal with the perfect data, e.g., the data are clean and balanced from distributional perspective, or have complete meta-information from structural perspective. To enhance their practicality, my long-term research goal is to improve the clustering methods in more challenging scenarios where the observed data are imperfect. Specifically, I plan to explore the correlation among the data distribution and structure, and design more accurate and more efficient clustering approaches on imperfect data.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Naoki-Kimura_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Naoki Kimura\" class=\"wp-image-708847\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Naoki-Kimura_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Naoki-Kimura_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"naoki-kimura-opens-in-new-tab\"><a href=\"http:\/\/nao-ki-mura.com\/\" target=\"_blank\" rel=\"noopener\">Naoki Kimura<\/a><\/h3>\n\n\n\n<p>The University of Tokyo<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/lab.rekimoto.org\/\" target=\"_blank\" rel=\"noopener\">Junichi Rekimoto<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Human-Computer Interaction, Wearable Computing, Pattern Recognition<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> I am developing an interface that allows every human being to use a computer without any special ability or training, like internal organs controlled by autonomic nerves. We still have many barriers to use computers. To use current computer systems, we need not only training, but also, at least vision and hands, or hearing and voice. The training keeps many people away from computers. Not all people have vision or hearing enough. Computers remain a tool for privileged people. For this goal, I have been working on Silent Speech Interfaces.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Tianyu-Pang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Tianyu Pang\" class=\"wp-image-708856\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Tianyu-Pang_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Tianyu-Pang_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"tianyu-pang-opens-in-new-tab\"><a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~tianyu\/\" target=\"_blank\" rel=\"noopener\">Tianyu Pang<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor:\u202f<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~jun\/index.shtml\" target=\"_blank\" rel=\"noopener\">Jun Zhu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Robust Machine Learning, Deep Generative Model<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research interests span the areas of machine learning and deep learning, especially trust-worthy machine learning which deals with the problem of adversarial examples and more general robustness. The adversarial examples are maliciously crafted with human imperceptible perturbations, but can easily fool high-performance deep neural networks to return wrong decisions. The goal of my research is to design reliable, general strategies to enhance the model robustness, and explain the phenomena of adversarial vulnerability by the concepts from generative modeling or casual inference.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Xu-Han_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Xu Han\" class=\"wp-image-708859\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Xu-Han_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Xu-Han_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"xu-han-opens-in-new-tab\"><a href=\"https:\/\/thucsthanxu13.github.io\/\" target=\"_blank\" rel=\"noopener\">Xu Han<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: Zhiyuan Liu, Maosong Sun<\/p>\n\n\n\n<p><b>Research interests:<\/b> Natural Language Processing, Information Extraction, Knowledge Graph<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research focuses on the intersection of natural language processing, information extraction, and knowledge graph. Up to now, various large-scale knowledge graphs have been constructed to organize world knowledge in a structured form, which can benefit many knowledge-driven applications. Although existing knowledge graphs may contain hundreds of millions of facts, they are still far away from completion and require further efforts to acquire new knowledge. The goal of my research is to extract knowledge from the text to form a complete knowledge graph and meanwhile utilize the knowledge graph to perform better language understanding.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Zili-Meng_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Zili Meng\" class=\"wp-image-708865\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Zili-Meng_170x210.jpg 170w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Zili-Meng_170x210-10x12.jpg 10w\" sizes=\"(max-width: 170px) 100vw, 170px\" \/><\/figure>\n\n\n\n<h3 id=\"zili-meng-opens-in-new-tab\"><a href=\"https:\/\/zilimeng.com\" target=\"_blank\" rel=\"noopener\">Zili Meng<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/routing.netlab.edu.cn\/tiki-index.php?page=Mingwei+Xu\" target=\"_blank\" rel=\"noopener\">Mingwei Xu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Networking Systems, Video Streaming, Deep Learning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> I conduct research in the area of computer networks, focusing on the intersection area between deep learning and networking system. When improving the performance of networking systems, existing deep learning techniques also introduce a series of drawbacks (e.g., heavyweight, noninterpretable, etc.). My long-term research goal is to make deep learning-based networking systems deployable in practice and enjoy the performance benefits without the drawbacks. Besides, I\u2019m also interested in the optimization of transport in video streaming with advanced techniques.<\/p>\n\n\n\n\n\n<p>Over one hundred distinguished PhD candidates from 31 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2019 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 12 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.<\/p>\n\n\n\n<div id=\"attachment_620817\" class=\"wp-caption aligncenter\" style=\"width: 1034px\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-620817 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-1024x576.jpg\" sizes=\"(max-width: 1024px) 100vw, 1024px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-1024x576.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-300x169.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-768x432.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-1066x600.jpg 1066w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-655x368.jpg 655w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-343x193.jpg 343w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-640x360.jpg 640w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-960x540.jpg 960w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-1280x720.jpg 1280w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788.jpg 1400w\" alt=\"2019 Microsoft Research Asia Fellows\" width=\"1024\" height=\"576\" aria-describedby=\"caption-attachment-620817\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-620817\" class=\"wp-caption-text\">Microsoft Research Asia 2019 fellows with Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&D Group, Microsoft Research Asia (first from right in back row), and Tie-Yan Liu, Assistant Managing Director of Microsoft Research Asia (first from left in back row).<\/p>\n<\/div>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Shijie-Cao_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Shijie Cao\" class=\"wp-image-620865\" \/><\/figure>\n\n\n\n<h3 id=\"shijie-cao\">Shijie Cao<\/h3>\n\n\n\n<p>Harbin Institute of Technology<\/p>\n\n\n\n<p>Supervisors: Lintao Zhang, Dechen Zhang<\/p>\n\n\n\n<p><b>Research interests:<\/b> Deep Learning, High Performance Computing, Reconfigurable Computing<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research interest lies at the intersection of deep learning and high-performance computing. Deep learning has revolutionized vision, language and speech fields with tremendous large and multi-layer deep neural networks. Inference on these bulky models often requires a huge amount of computational power. However, the slowing down of Moore\u2019s law and Dennard scaling has led to the end of rapid performance improvement in general-purpose processors. Running DNN models in a low-cost, low-latency and energy-ef\ufb01cient way is highly desirable and has attracted much attention in the research community. My long-term research goal is co-designing deep learning models\/algorithms and domain specific hardware architectures for efficient deep learning computing.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Zhenpeng-Chen_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Zhenpeng Chen\" class=\"wp-image-620874\" \/><\/figure>\n\n\n\n<h3 id=\"zhenpeng-chen-opens-in-new-tab\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/chenzhenpeng18.github.io\/\" target=\"_blank\" rel=\"noopener\">Zhenpeng Chen<\/a><\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/sei.pku.edu.cn\/~meih\/index_en.html\" target=\"_blank\" rel=\"noopener\">Hong Mei<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.liuxuanzhe.com\/\" target=\"_blank\" rel=\"noopener\">Xuanzhe Liu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Software Analytics, Web Mining<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> To help software practitioners understand their target audience and thus obtain actionable insights for improving user experience, powerful techniques, such as user profiling and sentiment analysis, have been proposed. However, many of them are performed based on user-generated texts, mainly English texts. Although some efforts have also been made with other languages, related techniques for non-English languages are far behind. This creates a considerable inequality in the quality of the related services received by non-English users, especially considering that about 75% of Internet users are non-English speakers. To alleviate such an inequality, the long-term goal of my research is to develop novel approaches to understanding users through ubiquitous data such as emojis. Since these data are language-independent, techniques powered by them can benefit users in different languages.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Yinpeng-Dong_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Yinpeng Dong\" class=\"wp-image-620871\" \/><\/figure>\n\n\n\n<h3 id=\"yinpeng-dong-opens-in-new-tab\"><a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~yinpeng\/\" target=\"_blank\" rel=\"noopener\">Yinpeng Dong<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~jun\/index.shtml\" target=\"_blank\" rel=\"noopener\">Jun Zhu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Robust Deep Learning, Adversarial Machine Learning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research focuses on the robustness issue of machine learning and deep learning models in the adversarial setting. These models are shown to be significantly vulnerable to adversarial examples, which are maliciously crafted by adding small perturbations to the normal ones, but make the target model produce erroneous predictions. The goal of my research is to thoroughly understand the robustness of these models through adversarial attacks; develop more robust machine learning and deep learning models, and build a comprehensive benchmark to evaluate adversarial robustness in various settings.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Insu-Han_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Insu Han\" class=\"wp-image-620859\" \/><\/figure>\n\n\n\n<h3 id=\"insu-han-opens-in-new-tab\"><a href=\"https:\/\/insuhan.github.io\/\" target=\"_blank\" rel=\"noopener\">Insu Han<\/a><\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/alinlab.kaist.ac.kr\/shin.html\" target=\"_blank\" rel=\"noopener\">Jinwoo Shin<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Large-scale Machine Learning, Approximate Algorithm<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> In many modern machine learning problems, it is desirable to develop scalable algorithms suitable for large-scale settings. My research goal is to design fast and reliable approximate algorithms for a wide range of machine learning problems. In particular, I am interested in problems related to matrix spectral functions, for instance, optimizing recommendation systems, inference of diverse subset selection and so on. For future research, I will focus on developing fast but theoretically guaranteed approximate algorithms in fundamental machine learning problems so that they are indeed applicable in practical settings.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Hirofumi-Inaguma_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Hirofumi Inaguma\" class=\"wp-image-620853\" \/><\/figure>\n\n\n\n<h3 id=\"hirofumi-inaguma-opens-in-new-tab\"><a href=\"https:\/\/hirofumi0810.github.io\/\" target=\"_blank\" rel=\"noopener\">Hirofumi Inaguma<\/a><\/h3>\n\n\n\n<p>Kyoto University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/sap.ist.i.kyoto-u.ac.jp\/members\/kawahara\/home_e.html\" target=\"_blank\" rel=\"noopener\">Tatsuya Kawahara<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Automatic speech recognition, speech translation<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Speech translation (ST) is a key technique to break the language barrier in human communication. Automatic speech recognition (ASR) and machine translation (MT) are its essential components. However, error propagations from the ASR module prevent the following MT module from translating source text correctly and this results in conveying incorrect information. To address this, end-to-end ST (E2E-ST) has been investigated to directly translate source speech to target languages. The long-term goal of my research is to explore novel approaches to realize the universal multilingual E2E-ST system, which takes contextual information shared among multiple languages into account and generates target translations in an online streaming manner.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Dahun-KIm_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Dahun KIm\" class=\"wp-image-620850\" \/><\/figure>\n\n\n\n<h3 id=\"dahun-kim-opens-in-new-tab\"><a href=\"https:\/\/sunghoonim.github.io\/\" target=\"_blank\" rel=\"noopener\">Dahun KIm<\/a><\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/rcv.kaist.ac.kr\/\" target=\"_blank\" rel=\"noopener\">In So Kweon<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Computer Vision and Machine Learning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> We are in the midst of content explosion, surrounded by an endless supply of images and videos. My research aims at creating machines that can understand the visual content with minimal human supervision. This also challenges the biggest bottleneck in recent deep learning technologies: having humans annotate a large set of data is expensive, error-prone, and limits the scalability to real-world applications. I have explored algorithms that can learn with weak or self-supervision. The ongoing and future directions aim for the video domain, including video editing and understanding tasks, where the limitation of human annotation is even more severe.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Chao_Liao_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Chao Liao\" class=\"wp-image-620880\" \/><\/figure>\n\n\n\n<h3 id=\"chao-liao-opens-in-new-tab\"><a href=\"http:\/\/chaoliao.org\/\" target=\"_blank\" rel=\"noopener\">Chao Liao<\/a><\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/itcs.shufe.edu.cn\/pinyan\/\" target=\"_blank\" rel=\"noopener\">Pingyan Lu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/apex.sjtu.edu.cn\/members\/yyu\" target=\"_blank\" rel=\"noopener\">Yong Yu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Theoretical Computer Science, Approximate Counting and Sampling<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Counting problems naturally arise in many subjects, including combinatorics, discrete probability, statistical physics and many others. From a computational perspective, we would like to obtain fast algorithms to count the number of certain objects (exactly or approximately), or prove that such algorithms do not exist. Over the past decades, many interesting results have been discovered. One highlight is the correspondence between the computational phase transition and the uniqueness of Gibbs measures on the hard-core model. Even though, our understanding of the whole picture is still far from complete. In the future, I plan to explore the connections between the existence of efficient counting algorithms and other subjects, hoping to provide new insights into the picture.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Muoi-Tran_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Muoi Tran\" class=\"wp-image-620862\" \/><\/figure>\n\n\n\n<h3 id=\"muoi-tran-opens-in-new-tab\"><a href=\"https:\/\/www.comp.nus.edu.sg\/~muoitran\/\" target=\"_blank\" rel=\"noopener\">Muoi Tran<\/a><\/h3>\n\n\n\n<p>National University of Singapore, School of Computing<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.comp.nus.edu.sg\/~kangms\/\" target=\"_blank\" rel=\"noopener\">Min Suk Kang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Network Security, Blockchain Security<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Whether it is a multibillion-dollar cryptocurrency (e.g., Bitcoin) or a toy blockchain implemented for fun, the agreements among distributed nodes heavily rely on the underlying peer-to-peer network for data transmission. Therefore, this network has been an attractive target of attacks in recent years\u2014large Internet Service Providers have been shown to be capable of disrupting it effectively and stealthily. Currently, we focus on both uncovering such a family of attacks and building practical detection systems against them. Our long-term research goal is to design secure peer-to-peer networks that fundamentally harden thousands of cryptocurrencies, and even against strong network adversary model such as Internet Service Providers.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Taining-Wang_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Taining Wang\" class=\"wp-image-620868\" \/><\/figure>\n\n\n\n<h3 id=\"taining-wang\">Taining Wang<\/h3>\n\n\n\n<p>National University of Singapore<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.comp.nus.edu.sg\/~chancy\/\" target=\"_blank\" rel=\"noopener\">Chee-Yong Chan<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Database query processing and optimization, data sampling, data analytics<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My long-term research goal is to develop algorithms to improve the query processing and data analytics functionalities in data management systems. Managing and analyzing data efficiently and effectively has always been a key concern in data management systems. However, there are still many unsolved problems in efficient and effective data processing. One example of such problems is on optimizing the execution order of the join operators in a join query. This is a critical problem in database query optimization, because a good join ordering can usually lead to orders of magnitude improvement on the query execution time. The main difficulty is to accurately estimate the size of the join query result with minimal computation and storage overhead. This is one of the open problems in efficient and effective data processing among others.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Chuhan-Wu_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Chuhan Wu\" class=\"wp-image-620847\" \/><\/figure>\n\n\n\n<h3 id=\"chuhan-wu-opens-in-new-tab\"><a href=\"https:\/\/wuch15.github.io\/\" target=\"_blank\" rel=\"noopener\">Chuhan Wu<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.tsinghua.edu.cn\/publish\/eeen\/3784\/2010\/20101219113838975735106\/20101219113838975735106_.html\" target=\"_blank\" rel=\"noopener\">Yongfeng Huang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Recommender Systems, User Modeling and Natural Language Processing<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> User modeling is important for understanding users and providing personalized services. However, existing user modeling methods usually rely on centralized storage of user behavior data and centralized model learning, which may bring a series of challenges on privacy, such as the privacy concerns of users, the difficulties to need data protect regulations and the risk of user data leakage. Thus, privacy-preserving user modeling is necessary. In the future, in addition to developing more accurate and efficient user modeling methods, I plan to explore how to model users and provide personalized services in a privacy-preserving manner.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Hongming-Zhang_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Hongming Zhang\" class=\"wp-image-620856\" \/><\/figure>\n\n\n\n<h3 id=\"hongming-zhang\">Hongming Zhang<\/h3>\n\n\n\n<p>The Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.cse.ust.hk\/~yqsong\/\" target=\"_blank\" rel=\"noopener\">Yangqiu Song<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Commonsense Reasoning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Commonsense reasoning, as one of the most challenging tasks in the natural language processing community, is crucial for machines to truly understand human language. In the future, I plan to focus on solving commonsense reasoning problems with structured knowledge. To be specific, I am trying to explore a structured way of representing commonsense knowledge, an efficient way of extracting commonsense knowledge, and a principle way of applying commonsense knowledge into downstream tasks.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Zizhao-Zhang_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Zizhao Zhang\" class=\"wp-image-620877\" \/><\/figure>\n\n\n\n<h3 id=\"zizhao-zhang-opens-in-new-tab\"><a href=\"https:\/\/zzz14.github.io\/\" target=\"_blank\" rel=\"noopener\">Zizhao Zhang<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/gaoyue.org\/cn_tsinghua\/people\/gaoyue_index.html\" target=\"_blank\" rel=\"noopener\">Yue Gao<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Brain Science, Graph Signal Processing, Complex Network<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> In recent years, many brain disorders have been found to be associated with topological structure abnormalities of large-scale functional networks, such as Alzheimer\u2019s (AD), Parkinson\u2019s (PD), autism spectrum disorder (ASD) and schizophrenia (SZ). However, human brain is a complex network and the brain functional activity is a real-time dynamic process. How to bridge the gap between the brain networks and the brain disorders is an important task. The long-term goal of my research is to establish the automated diagnosis and pathological mechanisms understanding of brain disorders using brain functional and structural network.<\/p>\n\n\n\n\n\n<p>Over one hundred distinguished PhD candidates from 40 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2018 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 11 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.<\/p>\n\n\n\n<div id=\"attachment_551808\" class=\"wp-caption aligncenter\" style=\"width: 1034px\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-551808 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/1-2018-Fellow-with-Peter-and-Hon-eyes-1024x513.jpg\" sizes=\"(max-width: 1024px) 100vw, 1024px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/1-2018-Fellow-with-Peter-and-Hon-eyes-1024x513.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/1-2018-Fellow-with-Peter-and-Hon-eyes-300x150.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/1-2018-Fellow-with-Peter-and-Hon-eyes-768x385.jpg 768w\" alt=\"Microsoft Research Asia 2018 fellows\" width=\"1024\" height=\"513\" aria-describedby=\"caption-attachment-551808\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-551808\" class=\"wp-caption-text\">Microsoft Research Asia 2018 fellows with Peter Lee, CVP, Microsoft Research (1st from the right); Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&D Group, Microsoft Research Asia (2nd from the left), and Lidong Zhou, Assistant Managing Director of Microsoft Research Asia (1st from the left).<\/p>\n<\/div>\n\n\n\n<div id=\"attachment_551811\" class=\"wp-caption aligncenter\" style=\"width: 1034px\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-551811 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/2-2018-Fellow-with-Raj-Reddy-1024x521.jpg\" sizes=\"(max-width: 1024px) 100vw, 1024px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/2-2018-Fellow-with-Raj-Reddy-1024x521.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/2-2018-Fellow-with-Raj-Reddy-300x153.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/2-2018-Fellow-with-Raj-Reddy-768x391.jpg 768w\" alt=\"Microsoft Research Asia 2018 fellows\" width=\"1024\" height=\"521\" aria-describedby=\"caption-attachment-551811\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-551811\" class=\"wp-caption-text\">After Award Ceremony, Microsoft Research Asia 2018 fellows had roundtable meeting with Raj Reddy, 1994 Turing Award Recipient.<\/p>\n<\/div>\n\n\n\n<div id=\"attachment_551814\" class=\"wp-caption aligncenter\" style=\"width: 1034px\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-551814 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/3-2018-Fellow-with-Yao-1024x683.jpg\" sizes=\"(max-width: 1024px) 100vw, 1024px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/3-2018-Fellow-with-Yao-1024x683.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/3-2018-Fellow-with-Yao-300x200.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/3-2018-Fellow-with-Yao-768x512.jpg 768w\" alt=\"Microsoft Research Asia 2018 fellows roundtable\" width=\"1024\" height=\"683\" aria-describedby=\"caption-attachment-551814\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-551814\" class=\"wp-caption-text\">Microsoft Research Asia 2018 fellows had roundtable meeting with Andrew Chi-Chih Yao, 2000 Turing Award Recipient.<\/p>\n<\/div>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng.png\" alt=\"Weiming Feng - Fellowships at Microsoft Research Asia\" class=\"wp-image-551820\" width=\"165\" height=\"165\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng.png 560w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng-150x150.png 150w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng-300x300.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng-180x180.png 180w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng-360x360.png 360w\" sizes=\"(max-width: 165px) 100vw, 165px\" \/><\/figure>\n\n\n\n<h3 id=\"weiming-feng-opens-in-new-tab\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/tcs.nju.edu.cn\/fengwm\/\" target=\"_blank\" rel=\"noopener\">Weiming Feng<\/a><\/h3>\n\n\n\n<p>Nanjing University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/tcs.nju.edu.cn\/yinyt\/Yitong_Yin_homepage.html\" target=\"_blank\" rel=\"noopener\">Yitong Yin<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Randomized Algorithms, Theory of Distributed Computing<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Sampling is an extensively studied topic in computer science and statistical physics. However, many classic and widely-used sampling algorithms such as single-site dynamics are fully adaptive and highly sequential. Nowadays, the size of data set grows rapidly so that the sequential sampling algorithms become inefficient. As a result, the distributed sampling algorithms attract more and more attention, especially on the area of distributed machine learning. However, the theory of distributed sampling is lacking of systematic studies. The long-term goal of my research is to propose novel algorithms and provide a better understanding of distributed sampling.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/2-Sunghoon-Im.png\" alt=\"Sunghoon Im - Fellowships at Microsoft Research Asia\" class=\"wp-image-551823\" width=\"163\" height=\"218\" \/><\/figure>\n\n\n\n<h3 id=\"sunghoon-im-opens-in-new-tab\"><a href=\"https:\/\/sunghoonim.github.io\/\">Sunghoon Im<\/a><\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/rcv.kaist.ac.kr\/index.php?mid=rcv_faculty\">In So Kweon<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests: <\/strong>Computer Vision and Machine Learning<\/p>\n\n\n\n<p><strong>Long-term research goal: <\/strong>3D scene understanding is essential for AR\/VR application and robotics application; however, capturing 3D data requires costly and bulky additional devices that reduce commercial use. My research goal is to provide a cost and size efficient solution for visual tracking and reconstruction in the hand-held device industry through the combination of modern geometric knowledge and machine learning techniques. In the future, I plan on exploring additional novel approaches that can <u>be applied<\/u> to commercial products.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/3-Dongup-Kwon.png\" alt=\"Dongup Kwon - Fellowships at Microsoft Research Asia\" class=\"wp-image-551826\" width=\"150\" height=\"200\" \/><\/figure>\n\n\n\n<h3 id=\"dongup-kwon-opens-in-new-tab\"><a href=\"https:\/\/hpcs.snu.ac.kr\/~dongup\/\" target=\"_blank\" rel=\"noopener\">Dongup Kwon<\/a><\/h3>\n\n\n\n<p>Seoul National University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/hpcs.snu.ac.kr\/~jangwoo\/\" target=\"_blank\" rel=\"noopener\">Jangwoo Kim<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Server Architecture, Accelerator Architecture, System Software<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> My research interests are in developing and evaluating computer systems targeting emerging server applications such as data analytics, machine learning, and brain simulations. Since such workloads demand much higher computational capabilities at unprecedented scales, it is critical to design extremely fast and scalable server systems. My research goal is to develop device-centric and extreme-scale server systems for large-scale and distributed workloads. I mainly focus on HW\/SW co-optimization and FPGA\/ASIC-based accelerators. I firmly believe that my research topic will have a significant impact on modern domain-specific or cloud-based computing paradigms.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li.png\" alt=\"Yikang Li - Fellowships at Microsoft Research Asia\" class=\"wp-image-551829\" width=\"199\" height=\"199\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li.png 440w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li-150x150.png 150w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li-300x300.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li-180x180.png 180w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li-360x360.png 360w\" sizes=\"(max-width: 199px) 100vw, 199px\" \/><\/figure>\n\n\n\n<h3 id=\"yikang-li-opens-in-new-tab\"><a href=\"http:\/\/cvboy.com\/\" target=\"_blank\" rel=\"noopener\">Yikang LI<\/a><\/h3>\n\n\n\n<p>The Chinese University of Hong Kong<\/p>\n\n\n\n<p>Supervisor: Xiaogang WANG<\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Deep Learning, Computer Vision, Vision and Language<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> With the development of the Computer Vision (CV) and Natural Language Processing (NLP), the intersection area, Vision and Language, has drawn increasingly more attentions these years. Integrating language with vision brings with it the possibility of expanding the horizons and tasks of the vision community. We have seen significant growth in image\/video-to-text tasks but many other potential applications of such integration \u2013 answering questions, dialog systems, and grounded language acquisition \u2013 remain largely unexplored. Going beyond such novel tasks, language can make a deeper contribution to vision: it provides a prism through which to understand the world. Therefore, my long-term goal is to make the computer Understanding and Express the visual world in a more human-friendly way by integrating the NLP research.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi.png\" alt=\"Jiaxin Shi - Fellowships at Microsoft Research Asia\" class=\"wp-image-551850\" width=\"184\" height=\"184\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi.png 430w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi-150x150.png 150w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi-300x300.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi-180x180.png 180w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi-360x360.png 360w\" sizes=\"(max-width: 184px) 100vw, 184px\" \/><\/figure>\n\n\n\n<h3 id=\"jiaxin-shi\">Jiaxin Shi<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~jun\/index.shtml\" target=\"_blank\" rel=\"noopener\">Jun Zhu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interest:<\/strong> Probabilistic Learning, Bayesian Inference, Bayesian Deep Learning<\/p>\n\n\n\n<p><strong>Long-term Research Goal:<\/strong> The main theme of my research is to address fundamental problems from probabilistic machine learning, Bayesian methods, and challenges from emerging fields like Bayesian deep learning. The goal is to increase the flexibility of probabilistic models, while keeping their inference and learning easy, and sufficiently scalable to solve real-world problems. Examples include unsupervised (semi-supervised) learning, structured prediction, and uncertainty estimation in supervised learning. I created and lead the development of ZhuSuan, a probabilistic programming library for Bayesian deep learning.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/6-Lili-Wei.png\" alt=\"Lili Wei - Fellowships at Microsoft Research Asia\" class=\"wp-image-551832\" width=\"162\" height=\"217\" \/><\/figure>\n\n\n\n<h3 id=\"lili-wei-opens-in-new-tab\"><a href=\"http:\/\/home.cse.ust.hk\/~lweiae\/\" target=\"_blank\" rel=\"noopener\">Lili WEI<\/a><\/h3>\n\n\n\n<p>The Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cse.ust.hk\/faculty\/scc\/\" target=\"_blank\" rel=\"noopener\">Shing-Chi Cheung<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Software engineering: program analysis and testing for mobile applications<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> The growing popularity of mobile devices has emerged the need of quality assurance for mobile applications. I have been focused on automatically identifying bugs in mobile applications. One of my major research projects is to characterize and detect compatibility issues induced by Android fragmentation. Due to its evolving and open nature, Android ecosystem is heavily fragmented. Various compatibility issues thus arise in Android applications. This has been well-recognized as one of the biggest challenges in Android application development. I have conducted studies to characterize and automatically detect these fragmentation-induced compatibility issues. In long term, I aim to develop a tool chain that can help app developers to automatically detect, diagnose and fix these compatibilities issues.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/7-Xingda-Wei.png\" alt=\"Xingda Wei - Fellowships at Microsoft Research Asia\" class=\"wp-image-551835\" width=\"187\" height=\"235\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/7-Xingda-Wei.png 261w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/7-Xingda-Wei-239x300.png 239w\" sizes=\"(max-width: 187px) 100vw, 187px\" \/><\/figure>\n\n\n\n<h3 id=\"xingda-wei\">Xingda Wei<\/h3>\n\n\n\n<p>Shanghai Jiaotong University<\/p>\n\n\n\n<p>Supervisor: Rong Chen, Haibo Chen and Bingyu Zang<\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Distributed systems and storage systems<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> Distributed transactional system is a key component for many data-center applications, which serves the storage backend for applications such as e-commerce applications, social networks and websites. However, efficient processing transactions are hard, especially in a distributed setting. My research is to improve the performance and reliability of distributed systems including distributed transactions. I mainly use two approaches to achieve this. The first is through an algorithmic way, which use algorithms to achieve better application properties (e.g. use fewer network communications) given common cases. The second is through hardware and system co-design, which integrate advanced hardware (like RDMA and HTM) into system designs. To this end, our innovations in distributed transactions can provide orders of magnitude better performance than priori solutions.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/8-Lijun-wu.png\" alt=\"Lijun Wu - Fellowships at Microsoft Research Asia\" class=\"wp-image-551838\" width=\"198\" height=\"198\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/8-Lijun-wu.png 328w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/8-Lijun-wu-150x150.png 150w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/8-Lijun-wu-300x300.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/8-Lijun-wu-180x180.png 180w\" sizes=\"(max-width: 198px) 100vw, 198px\" \/><\/figure>\n\n\n\n<h3 id=\"lijun-wu\">Lijun Wu<\/h3>\n\n\n\n<p>Sun Yat-sen University<\/p>\n\n\n\n<p>Supervisor: Jianhuang Lai and Tie-Yan Liu<\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Machine Learning, Reinforcement Learning, Neural Machine Translation<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> Neural machine translation (NMT) has drawn more and more attention in both academia and industry. With the development of deep learning and strong machine learning methods, NMT has achieved near human-level performance in some language pairs. However, current state-of-the-art NMT models require large amounts of parameters and still take several days to achieve a high performance. Besides, the fixed model structure and loss function make the NMT learning problem not easy to make a big breakout. My research goal is to design light and efficient machine learning models for machine translation, and eventually try to automatically \u201clearning to teach\u201d the optimal model for NMT or other NLP tasks.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu.png\" alt=\"Fengli Xu - Fellowships at Microsoft Research Asia\" class=\"wp-image-551841\" width=\"195\" height=\"195\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu.png 373w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu-150x150.png 150w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu-300x300.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu-180x180.png 180w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu-360x360.png 360w\" sizes=\"(max-width: 195px) 100vw, 195px\" \/><\/figure>\n\n\n\n<h3 id=\"fengli-xu-opens-in-new-tab\"><a href=\"http:\/\/fenglixu.com\/\" target=\"_blank\" rel=\"noopener\">Fengli Xu<\/a><\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/fi.ee.tsinghua.edu.cn\/~liyong\/\" target=\"_blank\" rel=\"noopener\">Yong Li<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Data-driven Human Behavior Modelling and Data Privacy<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> The dramatic proliferation of smart devices has significantly contributed to the increasingly available fine-grained human behavioral big data. Such datasets have enormous potential in pushing forward the frontier of human behavior modelling, but they also pose challenges to current analytic frameworks and user privacy preservation techniques. My research contributes to the endeavor of harnessing the power of human behavioral big data in two ways: developing compatible analytic frameworks and proposing privacy models that allow datasets to safely circulate. My long term research goal is to expand my current research to facilitate advanced privacy-preserving human-centric computing.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/10-Pan-Zhou.png\" alt=\"Pan Zhou - Fellowships at Microsoft Research Asia\" class=\"wp-image-551844\" width=\"153\" height=\"222\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/10-Pan-Zhou.png 226w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/10-Pan-Zhou-207x300.png 207w\" sizes=\"(max-width: 153px) 100vw, 153px\" \/><\/figure>\n\n\n\n<h3 id=\"pan-zhou\">Pan Zhou<\/h3>\n\n\n\n<p>National University of Singapore<\/p>\n\n\n\n<p>Supervisor: Jiashi Feng and Shuicheng Yan<\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Structural Data Analysis, Optimization Algorithm, Deep Learning Theory<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> Data modeling method and optimization algorithm are two key factors for achieving success in real applications. For instance, thanks to the strong data modeling ability of deep learning and the effectiveness of stochastic backpropagation algorithm, we have achieved remarkable success in many AI fields. So like my previous work, in the future I continue to focus my research interest on developing stronger data modeling methods and more efficient optimization algorithms, as well as establishing their theoretical performance guarantees. Especially, I will put more my efforts on designing effective network architecture and developing more efficient training algorithm in deep learning because of its favorable usage in applications. My future goal is to provide a new tool, including a data modeling technique and its algorithm, for better modeling the real data with high efficiency.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/11-Xizhou-Zhu.png\" alt=\"Xizhou Zhu - Fellowships at Microsoft Research Asia\" class=\"wp-image-551847\" width=\"156\" height=\"229\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/11-Xizhou-Zhu.png 218w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/11-Xizhou-Zhu-204x300.png 204w\" sizes=\"(max-width: 156px) 100vw, 156px\" \/><\/figure>\n\n\n\n<h3 id=\"xizhou-zhu\">Xizhou Zhu<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisor: <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/bainguo\/\">Baining Guo<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> and <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/staff.ustc.edu.cn\/~xjchen99\/\" target=\"_blank\" rel=\"noopener\">Xuejin Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><strong>Research interests:<\/strong> Deep Learning, Video Recognition<\/p>\n\n\n\n<p><strong>Long-term research goal:<\/strong> Recent years have witnessed significant success of deep convolutional neutral networks for image recognition. With their success, the recognition tasks have been extended from image domain to video domain. Fast and accurate video recognition is crucial for high-value scenarios, e.g., autonomous driving and video surveillance. I have proposed flow-based methods to exploit motion for video recognition tasks, which has achieved both faster speed and better accuracy. These methods share the same principles: motion estimation module built into the network architecture and end-to-end learning of all modules performed over multiple frames. I insist on principled multi-frame feature learning instead of heuristics, which is general for different tasks. For future research, I will focus on not only better speed-accuracy tradeoff but also challenges beyond speed and accuracy for video recognition, e.g. low latency and stability.<\/p>\n\n\n\n\n\n<h2 id=\"2017-fellows-announced\">2017 fellows announced<\/h2>\n\n\n\n<p>More than 100 distinguished PhD candidates from 38 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore have applied for the 2017 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 10 outstanding PhD candidates whose exceptional talent and innovation in computer science\u2013related research identifies them as emerging leaders in the Asia-Pacific region.<\/p>\n\n\n\n<div id=\"attachment_435546\" class=\"wp-caption aligncenter\" style=\"width: 818px\"><img loading=\"lazy\" decoding=\"async\" class=\" wp-image-435546\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017_MSRA_Fellows_Peter_Hon.jpg\" sizes=\"(max-width: 808px) 100vw, 808px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017_MSRA_Fellows_Peter_Hon.jpg 1616w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017_MSRA_Fellows_Peter_Hon-300x164.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017_MSRA_Fellows_Peter_Hon-768x421.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017_MSRA_Fellows_Peter_Hon-1024x561.jpg 1024w\" alt=\"Microsoft Research Asia fellows 2017\" width=\"808\" height=\"442\" aria-describedby=\"caption-attachment-435546\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-435546\" class=\"wp-caption-text\">Microsoft Research Asia 2017 fellows with Peter Lee, CVP, Microsoft Research (far right); Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&D Group, Microsoft Research Asia (second from the left); and Tim Pan, Senior Outreach Director of Microsoft Research Asia (far left).<\/p>\n<\/div>\n\n\n\n<div id=\"attachment_435543\" class=\"wp-caption aligncenter\" style=\"width: 818px\"><img loading=\"lazy\" decoding=\"async\" class=\" wp-image-435543\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017-Fellow-with-John-Hopcroft.jpg\" sizes=\"(max-width: 808px) 100vw, 808px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017-Fellow-with-John-Hopcroft.jpg 1616w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017-Fellow-with-John-Hopcroft-300x165.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017-Fellow-with-John-Hopcroft-768x422.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/2017-Fellow-with-John-Hopcroft-1024x563.jpg 1024w\" alt=\"Microsoft Research Asia fellows 2017\" width=\"808\" height=\"444\" aria-describedby=\"caption-attachment-435543\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-435543\" class=\"wp-caption-text\">After the award ceremony, Microsoft Research Asia 2017 fellows had roundtable meeting with John Hopcroft, 1986 Turing Award Recipient.<\/p>\n<\/div>\n\n\n\n<p>Microsoft Research Asia recognizes the following fellows, who represent the best and the brightest PhD candidates in the Asia-Pacific region.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Seung-Hwan-Baek.jpg\" alt=\"Seung-Hwan Baek Microsoft Research Asia fellow\" class=\"wp-image-435561\" \/><\/figure>\n\n\n\n<h3 id=\"seung-hwan-baek-opens-in-new-tab\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/sites.google.com\/view\/shbaek\/\" target=\"_blank\" rel=\"noopener\">Seung-Hwan Baek<\/a><\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/vclab.kaist.ac.kr\/minhkim\/index.html\" target=\"_blank\" rel=\"noopener noreferrer\">Min H. Kim<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Computational imaging and computational photography<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Visual data is key to understanding the real world for both humans and robots. However, due to the limitations of current imaging systems, capturing the real world in full fidelity is still challenging. My research goal is to overcome these limitations by designing computational imaging systems based on computer graphics, computer vision, and optics. I take a multidisciplinary approach, not only designing a problem in a mathematical form, but also exploring diverse optical phenomena to reveal invisible properties of the real world: depth, material reflectance, and spectrum. I envision that my research will provide new tools for better understanding the real world, which is essential for many applications such as VR\/AR and autonomous driving.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Yue-Cao.jpg\" alt=\"Yue Cao Microsoft Research Asia fellow\" class=\"wp-image-435570\" \/><\/figure>\n\n\n\n<h3 id=\"yue-cao\">Yue Cao<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: Jianmin Wang and Mingsheng Long<\/p>\n\n\n\n<p><b>Research interests:<\/b> Deep learning and computer vision<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Due to the tremendous increase in the size of data with high dimensions, it remains a great challenge to efficiently search for multimedia data. To guarantee both retrieval quality and computation efficiency, the approximate nearest neighbor (ANN) search has attracted increasing attention and is shown to be very useful for many practical problems, such as similar image search, object retrieval, cross-modal retrieval, and more. In the future, I\u2019ll continue to explore novel approaches to further improve the accuracy and efficiency of approximate nearest neighbor search.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Bojie-Li.jpg\" alt=\"Bojie Li Microsoft Research Asia fellow\" class=\"wp-image-435549\" \/><\/figure>\n\n\n\n<h3 id=\"bojie-li\">Bojie Li<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors: Enhong Chen and Lintao Zhang<\/p>\n\n\n\n<p><b>Research interests:<\/b> Networked systems and reconfigurable hardware<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> I\u2019m pushing forward research in accelerating datacenter infrastructure through reconfigurable hardware. There is an increasing performance mismatch between CPU and specialized accelerators (such as GPU, TPU), storage, and network components. Because Microsoft is deploying reconfigurable hardware in the public cloud, it gives me the opportunity to explore this potentially fruitful research area, now in its infancy. Our solution is to offload common patterns in computation, communication, and coordination from CPU to programmable NIC, as well as rethinking the architecture and programming model for the network switch, NIC, OS, and applications. To this end, we have designed systems to achieve significant speedup in network functions, in-memory key-value store, distributed transactions, and container networking. I hope to glue these pieces of work into a coherent picture of datacenter infrastructure, and ultimately arrive at elementary abstractions to build efficient and scalable systems in the future.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Chongxuan-Li.jpg\" alt=\"Chongxuan Li Microsoft Research Asia fellow\" class=\"wp-image-435555\" \/><\/figure>\n\n\n\n<h3 id=\"chongxuan-li\">Chongxuan Li<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: Bo Zhang and Jun Zhu<\/p>\n\n\n\n<p><b>Research interests:<\/b> Deep learning and deep generative models<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research interests are primarily on deep generative models (DGMs), which conjoin the flexibility of deep neural networks and the inference power of generative approaches. Via probabilistic modelling and inference, DGMs can handle the uncertainty of the input data and extract meaningful features without supervision, which are key to building human-like AI. My long-term research goal is to develop novel DGMs for challenging learning tasks, especially semi-supervised DGMs on partially labelled data, structured DGMs on sequential data, and DGMs with decision making in reinforcement learning settings.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Pengfei-Liu.jpg\" alt=\"Pengfei Liu Microsoft Research Asia fellow\" class=\"wp-image-435555\" \/><\/figure>\n\n\n\n<h3 id=\"pengfei-liu\">Pengfei Liu<\/h3>\n\n\n\n<p>Fudan University<\/p>\n\n\n\n<p>Supervisors: Xuanjing Huang and Xipeng Qiu<\/p>\n\n\n\n<p><b>Research interests:<\/b> Natural language processing and deep learning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Deep learning methods have shown intriguing opportunities for natural language processing (NLP). Many typical tasks can be handled using end-to-end neural network\u2013based models. However, it\u2019s still a challenging problem to employ the correlations among different tasks and design a unified architecture to deal with different task simultaneously. So, my research work focuses on deep learning and multitask learning in NLP, hoping that a more general framework can be used to handle various tasks.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Xuan-Lu.jpg\" alt=\"Xuan Lu Microsoft Research Asia fellow\" class=\"wp-image-435555\" \/><\/figure>\n\n\n\n<h3 id=\"xuan-lu-opens-in-new-tab\"><a href=\"http:\/\/sei.pku.edu.cn\/~luxuan13\/\" target=\"_blank\" rel=\"noopener\">Xuan Lu<\/a><\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/sei.pku.edu.cn\/~meih\/index_en.html\" target=\"_blank\" rel=\"noopener\">Hong Mei<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> and Xuanzhe Liu<\/p>\n\n\n\n<p><b>Research interests:<\/b> Software analytics for mobile computing systems and applications<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> The wide adoption of smartphones and tablet computers has triggered a surge of developing mobile applications (apps) in recent years. App stores, such as Apple Store and Google Play, continuously attract millions of developers to release and popularize their apps. At the same time, the competition among apps is increasingly fierce. In this context, developers are eager to know how users behave in apps and how they evaluate the app in app stores, so that they can better understand users and find problems in their apps to tackle. The long-term goal of my research is to provide directive suggestions to developers to optimize mobile apps based on user behavior analysis.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Chang-Hyun-Park.jpg\" alt=\"Chang Hyun Park Microsoft Research Asia fellow\" class=\"wp-image-435552\" \/><\/figure>\n\n\n\n<h3 id=\"chang-hyun-park-opens-in-new-tab\"><a href=\"https:\/\/iamchanghyunpark.github.io\/\" target=\"_blank\" rel=\"noopener\">Chang Hyun Park<\/a><\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/calab.kaist.ac.kr\/~jhuh\/\" target=\"_blank\" rel=\"noopener\">Jaehyuk Huh<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Virtual memory and the underlying memory hierarchy<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research interests are focused toward the computer architecture and system software stack of computing. Designing computer systems cannot be driven exclusively by either computer architects or system software developers. The two groups need to work hand-in-hand to design efficient systems. My research is focused on the memory systems of both hardware and software. My research interests and goals include improving virtual memory systems, which are increasing in how much they are relied on. My goal is to help prepare virtual memory systems for such a future.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Zhaofan-Qiu.jpg\" alt=\"Zhaofan Qiu Microsoft Research Asia fellow\" class=\"wp-image-435576\" \/><\/figure>\n\n\n\n<h3 id=\"zhaofan-qiu\">Zhaofan Qiu<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors: Xinmei Tian and Tao Mei<\/p>\n\n\n\n<p><b>Research interests:<\/b> Deep learning, video understanding, and multimedia analysis<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My previous research mainly focused on building video understanding systems based on deep learning techniques. Following this direction, I will further explore the new possibilities of video understanding. The first dimension is to build an advanced video understanding system with capabilities such as action localization and action detection. Unlike traditional action recognition tasks, action localization and detection need to find when and where the action happens, which is much more challenging. The second dimension is to identify more powerful video representation approaches. One of the possible solutions is to adapt the image representation to the video domain, as well as training a very deep 3D CNN or deep RNN to improve the performance of current video understanding systems.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Yu-Wu.jpg\" alt=\"Yu Wu Microsoft Research Asia fellow\" class=\"wp-image-435567\" \/><\/figure>\n\n\n\n<h3 id=\"yu-wu\">Yu Wu<\/h3>\n\n\n\n<p>Beihang University<\/p>\n\n\n\n<p>Supervisors: Zhoujun Li and <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/mingzhou\/\" target=\"_blank\" rel=\"noopener\">Ming Zhou<\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Dialog systems and natural language processing<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Chatbots are bringing us to a new technology era, the era of conversational interface. It is an era that won\u2019t require a mouse or a keyboard, where human and machine can communicate with each other through conversations. My research goal is to develop intelligent chatbots that are able to communicate with humans for various topics, using the large amount of real conversation data on the Internet. My current research focuses on context-aware chatbots, and achieves significant improvement in contrast to conventional approaches. In the future, my research will focus on two aspects: 1) Let chatbots interact with human proactively. 2) Integrate prior knowledge into chatbots.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2017\/10\/Yuxiang-Yang.jpg\" alt=\"Yuxiang Yang Microsoft Research Asia fellow\" class=\"wp-image-435573\" \/><\/figure>\n\n\n\n<h3 id=\"yuxiang-yang\">Yuxiang Yang<\/h3>\n\n\n\n<p>University of Hong Kong<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/qift.weebly.com\/giulio-chiribella.html\/\" target=\"_blank\" rel=\"noopener\">Giulio Chiribella<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Quantum information and computation<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> The past few years witnessed both a rapid development of quantum technologies and new challenges in implementing long-distance quantum communication and long-term storage of quantum bits. My research focuses on implementing efficient quantum protocols and algorithms, which are tailor-made to reduce the consumption of quantum resources. I have been working on compression protocols for probes generated by quantum sensors and quantum algorithms with the minimum energy consumption. For the next step, I will focus on incorporating these protocols and algorithms in complex networks of quantum devices, aiming at achieving the quantum advantage with limited quantum resources.<\/p>\n\n\n\n\n\n<h2 id=\"2016-fellows-announced\">2016 fellows announced<\/h2>\n\n\n\n<p>Over one hundred distinguished PhD candidates from 40 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2016 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 10 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.<\/p>\n\n\n\n<div id=\"attachment_318728\" class=\"wp-caption aligncenter\" style=\"width: 820px\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-318728\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-on-stage-3.jpg\" sizes=\"(max-width: 810px) 100vw, 810px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-on-stage-3.jpg 5319w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-on-stage-3-300x169.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-on-stage-3-768x434.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-on-stage-3-1024x579.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-on-stage-3-343x193.jpg 343w\" alt=\"Microsoft Research Asia fellows 2016\" width=\"810\" height=\"458\" aria-describedby=\"caption-attachment-318728\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-318728\" class=\"wp-caption-text\">Microsoft Research Asia 2016 fellows with Peter Lee, CVP, Microsoft Research (far right); Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&D Group, Microsoft Research Asia (second from the left), and Baining Guo, Assistant Managing Director of Microsoft Research Asia (far left).<\/p>\n<\/div>\n\n\n\n<div id=\"attachment_318740\" class=\"wp-caption aligncenter\" style=\"width: 818px\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-318740\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-with-Adi-2.jpg\" sizes=\"(max-width: 808px) 100vw, 808px\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-with-Adi-2.jpg 3874w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-with-Adi-2-300x170.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-with-Adi-2-768x434.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-with-Adi-2-1024x579.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/all-with-Adi-2-343x193.jpg 343w\" alt=\"Microsoft Research Asia fellows 2016\" width=\"808\" height=\"456\" aria-describedby=\"caption-attachment-318740\">\n<p>&nbsp;<\/p>\n<p id=\"caption-attachment-318740\" class=\"wp-caption-text\">After Award Ceremony, Microsoft Research Asia 2016 fellows had roundtable meeting with Adi Shamir, 2002 Turing Award Recipient.<\/p>\n<\/div>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Chuang-Gan-1-e1478654999848.jpg\" alt=\"Chuang Gan Microsoft Research Asia fellow\" class=\"wp-image-318527\" \/><\/figure>\n\n\n\n<h3 id=\"chuang-gan\">Chuang Gan<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/itcs.tsinghua.edu.cn\/yao\/\" target=\"_blank\" rel=\"noopener noreferrer\">Andrew Chi-Chih Yao<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Deep Learning, Computer Vision, and Multimedia<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> The increasing ubiquity of devices capable of capturing videos has led to an explosion in the amount of recorded video content. Instead of \u201ceyeballing\u201d the videos for potential useful information, it is desirable to develop automatic video analysis and understanding algorithms. My research focuses on deep learning and its application to large-scale video understanding, e.g. video action recognition, multimedia event detection, video captioning, etc. Developing new algorithms to teach machines to understand the video content is my long-term research goal.<\/p>\n\n\n\n<h3 id=\"hyeonwoo-noh-opens-in-new-tab\"><a href=\"http:\/\/cvlab.postech.ac.kr\/~hyeonwoonoh\/\" target=\"_blank\" rel=\"noopener noreferrer\">Hyeonwoo Noh<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/cvlab.postech.ac.kr\/~hyeonwoonoh\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Hyeonwoo-Noh-e1478655129277.jpg\" alt=\"Hyeonwoo Noh Microsoft Research Asia Fellow\" class=\"wp-image-318530\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>Pohang University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/cvlab.postech.ac.kr\/~bhhan\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bohyung Han<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Visual understanding based on natural language<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research objective is building visual understanding algorithm based on natural language. Contrary to defining different problem setting for different types of visual understanding, using natural language as an interface for a visual understanding system provides a unified testbed for various visual understanding abilities. Solving multiple visual understanding problems with single algorithm based on natural language based interface invokes various research questions such as executing appropriate understanding function for different language based queries or sharing common visual concepts across multiple understanding functions. I believe finding solutions to these challenges is key to building generally applicable visual understanding algorithms.<\/p>\n\n\n\n<h3 id=\"jie-zhang-opens-in-new-tab\"><a href=\"https:\/\/sites.google.com\/site\/zhangjiemarina\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jie Zhang<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/sites.google.com\/site\/zhangjiemarina\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Jie-Zhang-e1478655175391.jpg\" alt=\"Jie Zhang Microsoft Research Asia fellow\" class=\"wp-image-318533\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/sei.pku.edu.cn\/~zhanglu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Lu Zhang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Software testing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>When ensuring software quality, traditionally developers\/testers explore program behaviors through test augmentation, either manually or with the aid of automatic test generation tools. However, the current test augmentation approaches have limitations, due to which many software systems are put into service with many behaviors unexplored, incurring serious threats to software quality. To complement the traditional approaches and better explore the program under test, I propose Variation Deduction, which is a new methodology that aims to explore more program behaviors through generating program variants instead of augmenting tests. In my future work, I plan to find valuable variation operators and deduction constraints, and make Variation Deduction applicable and valuable.<\/p>\n\n\n\n<h3 id=\"jun-nishida-opens-in-new-tab\"><a href=\"http:\/\/junnishida.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jun Nishida<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/junnishida.net\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Jun-Nishida-e1478655211107.jpg\" alt=\"Jun Nishida Microsoft Research Asia fellow\" class=\"wp-image-318536\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>University of Tsukuba<\/p>\n\n\n\n<p>Supervisor: Kenji Suzuki<\/p>\n\n\n\n<p><b>Research interests: <\/b>Augmenting Social and Embodied Experiences among People by Wearable Devices, Human-Computer Interaction, Biosignal Processing, Virtual Reality, Haptics<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research goal is to develop and establish new techniques and style for interpersonal communication that facilitate social and embodied interactions among people. Especially, I focus on devices that allow people to share and reproduce one\u2019s sensory and kinesthetic experiences to achieve natural communication in rehabilitations and design process. We have been proposing 1) a wearable suit that transforms a wearer\u2019s embodiment into that of a child for architectural design procedure, and 2) a paired wearable device that can share muscle activity bi-directionally for assisting rehabilitations. The achievement contributes to the technology for augmented human and benefits from concrete applications which reveals new aspects of human behaviors.<\/p>\n\n\n\n<h3 id=\"li-chen-opens-in-new-tab\"><a href=\"http:\/\/home.cse.ust.hk\/~lchenad\/\" target=\"_blank\" rel=\"noopener noreferrer\">Li Chen<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/home.cse.ust.hk\/~lchenad\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Li-Chen-e1478655247357.jpg\" alt=\"Li Chen Microsoft Research Asia fellow\" class=\"wp-image-318539\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>The Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cse.ust.hk\/~kaichen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kai Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Computing infrastructure & Datacenter networks<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>In this era of Big Data, both the volume of the data and the complexity of models to make sense of data are growing rapidly. However, speed of commodity processor and the limited memory size of commodity server force applications to scale out to large clusters. Most of the recent interesting applications, such as web search, recommendation systems, and deep networks, run on clusters of thousands of machines for both small companies and large enterprises. My long term research goal is accelerating application performance in this large-scale computing setting, especially the datacenter networks, as I believe this area presents challenging problems that can shape the future computing stack. My main research projects have been along this line: designing efficient networking for large-scale DC applications, with focus on flow scheduling for distributed and parallel computing systems.<\/p>\n\n\n\n<p><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/comp.nus.edu.sg\/~loiluu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Loi LUU<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/comp.nus.edu.sg\/~loiluu\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Loi-LUU-e1478655282616.jpg\" alt=\"Loi LUU Microsoft Research Asia fellow\" class=\"wp-image-318542\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>National University of Singapore<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/comp.nus.edu.sg\/~prateeks\/\" target=\"_blank\" rel=\"noopener noreferrer\">Prateek Saxena<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Blockchain-based Cryptocurrencies and Distributed Consensus Algorithms<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Blockchain-based cryptocurrencies, such as Bitcoin and 250 similar alt-coins, embody at their core a blockchain protocol \u2014 a mechanism for a distributed network of computational nodes to periodically agree on a set of new transactions. Designing a secure blockchain protocol is very challenging because often the protocol is run in open networks with presence of adversary and without any trusted PKI, or nodes do not have inherent identities. My long-term research goal is to improve the security and scalability of these agreement protocols. To achieve this goal, my current researches focus on three main directions. The first direction is to study and understand the current design of existing blockchain systems like Bitcoin or Ethereum. The main and most important piece of my research is to design secure and scalable blockchain protocols, given all the knowledge and analyses from existing blockchain systems. The third and interesting research direction is to devise new use-cases for blockchains and improve the usability of blockchain protocols.<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Qiang-Liu-e1478655356343.jpg\" alt=\"Qiang Liu Microsoft Research Asia fellow\" class=\"wp-image-318545\" \/><\/figure>\n\n\n\n<h3 id=\"qiang-liu\">Qiang Liu<\/h3>\n\n\n\n<p>Chinese Academy of Sciences<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/nlpr-web.ia.ac.cn\/english\/irds\/People\/lwang\/lwang.html\" target=\"_blank\" rel=\"noopener noreferrer\">Liang Wang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Data mining, user modeling, recommender systems, deep learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research focuses on user modeling, i.e., how to describe a user based on his or her online behavioral logs and generated contents. This topic has two parts. The first part is normal user modeling, which utilizes users\u2019 daily data to analyze their interests and demands, and predict what they would like to do next. The long-term goal of this part of research is to model a variety of contextual information and multimodal contents in daily applications for learning better user representations. The second part is abnormal user modeling, which utilizes patterns in normal data to detect abnormal behaviors, suspicious users, as well as misinformation on the web for security propose. The long-term goal of this part of research is to learn representations from large-scale daily data on the web, and make unsupervised detection and judgement of abnormal data.<\/p>\n\n\n\n<h3 id=\"wenhao-li-opens-in-new-tab\"><a href=\"http:\/\/ipads.se.sjtu.edu.cn\/doku.php?id=pub:members:wenhao_li\" target=\"_blank\" rel=\"noopener noreferrer\">Wenhao Li<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ipads.se.sjtu.edu.cn\/doku.php?id=pub:members:wenhao_li\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Wenhao-Li-e1478655368274.jpg\" alt=\"Wenhao Li Microsoft Research Asia fellow\" class=\"wp-image-318548\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ipads.se.sjtu.edu.cn\/doku.php?id=pub:members:haibo_chen\" target=\"_blank\" rel=\"noopener noreferrer\">Haibo Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> OS and Architecture, System and Mobile Security<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research goal is to bring systems with a good balance of performance, security and convenience to the real-world users. The approach I take is to improving the security and dependability of modern systems with practical hardware and software technologies. I am particularly interested in hardware-assisted security solutions which the industry is likely to adopt. I am currently exploring various forms of hardware-assisted and software designs in two aspects. The first is leveraging new hardware features (ARM TrustZone, Intel SGX, Intel VT, etc.) to improve the dependability and security of current mobile and cloud systems. The second is building and optimizing high performance platforms that benefit from system security.<\/p>\n\n\n\n<h3 id=\"yingce-xia-opens-in-new-tab\"><a href=\"http:\/\/home.ustc.edu.cn\/~xiayingc\/\" target=\"_blank\" rel=\"noopener noreferrer\">Yingce Xia<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/home.ustc.edu.cn\/~xiayingc\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Yingce-Xia-e1478655410780.jpg\" alt=\"Yingce Xia Microsoft Research Asia fellow\" class=\"wp-image-318551\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/staff.ustc.edu.cn\/~ynh\/\" target=\"_blank\" rel=\"noopener noreferrer\">Nenghai Yu<\/a> and <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/tyliu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Tie-Yan Liu<\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Machine Learning, Artificial Intelligence<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>While deep learning is making surprising progress in past few years, large amounts of labeled samples are needed for the training procedure. However, human labeling is usually very costly, leading to limited training data scale. To overcome such the difficulty, in our previous work, we propose the dual learning framework that can efficiently leverage unlabeled data to overcome labeled data shortage. We take neural machine translation (NMT) as the first step and achieve significant improvements. Towards this end, my long-term research goal is to fulfill such a dual learning framework, which mainly includes two aspects: 1) Enrich dual learning with more techniques and further improve NMT performances based on the enriched framework; 2) Apply dual learning to other areas like speech recognition\/text to speech, image caption\/image generation and so on.<\/p>\n\n\n\n<h3 id=\"ziqiang-cao-opens-in-new-tab\"><a href=\"http:\/\/www4.comp.polyu.edu.hk\/~cszqcao\/\" target=\"_blank\" rel=\"noopener noreferrer\">Ziqiang Cao<\/a><\/h3>\n\n\n\n<figure class=\"wp-block-image alignleft\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www4.comp.polyu.edu.hk\/~cszqcao\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img loading=\"lazy\" decoding=\"async\" width=\"170\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Ziqiang-Cao-e1478655438806.jpg\" alt=\"Ziqiang Cao\" class=\"wp-image-318554\" \/><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/figure>\n\n\n\n<p>The Hong Kong Polytechnic University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www4.comp.polyu.edu.hk\/~cswjli\/\" target=\"_blank\" rel=\"noopener noreferrer\">Wenjie Li<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Natural language processing, deep learning<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> The application of deep learning is growing fashionable in natural language processing. However, the generated features seldom have the explicit meaning. Therefore, my research goal is to develop both effective and understandable techniques for natural language processing, especially automatic summarization. To this end, I plan to make full use of the traditional research outputs in natural language processing. I am interested in the combination of deep learning and prior knowledge to better bridge the gap between natural language and structured data.<\/p>\n\n\n\n\n\n<h3 id=\"wei-bai\">Wei Bai<\/h3>\n\n\n\n<p>The Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cse.ust.hk\/~kaichen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kai Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Data center networking<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> Many data centers have been built around the world to provide various services and applications. In data centers, hundreds of thousands of servers are interconnected by using data center networks. My long-term research goal is to improve the performance of data center networks. To achieve this goal, I am focusing on designing new transport mechanisms to satisfy the low latency requirements of today\u2019s cloud applications. In the future, I plan to continue working on data center transport on two fronts: designing new solutions for multi-queue multi-service data centers as well as improving network performance for large-scale Remote Direct Memory Access deployments.<\/p>\n\n\n\n<h3 id=\"dong-deng\">Dong Deng<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/dbgroup.cs.tsinghua.edu.cn\/fengjh\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jianhua Feng<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/dbgroup.cs.tsinghua.edu.cn\/ligl\/\" target=\"_blank\" rel=\"noopener noreferrer\">Guoliang Li<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Database, data management, data cleaning, and data integration<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research focuses on data cleaning, i.e., how to deal with errors and inconsistencies in information systems. As an example, in many applications such as data integration, commercial organizations need to collect data from various sources to conduct analysis and make decisions. The data from these different sources typically contain inconsistencies. For example, the same person\u2019s name may be misspelled or have different formats, such as with or without middle name. Such inconsistencies make it more challenging to link records from different places and answer queries approximately. My long-term research goal is to develop algorithms in order to make query answering and information retrieval efficient in the presence of such inconsistencies and errors.<\/p>\n\n\n\n<h3 id=\"li-dong\">Li Dong<\/h3>\n\n\n\n<p>Beihang University<\/p>\n\n\n\n<p>Supervisors: Wei Li, Ke Xu<\/p>\n\n\n\n<p><b>Research interests: <\/b>Natural language processing, artificial intelligence<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b> My research goal is to develop novel and effective techniques for natural language processing, especially question ansering. In order to build better natural language interfaces for machines, we need to learn to sufficiently represent text from large volumes of data, and bridge the gap between natural language and structured data. I am interested in leveraging extensive unannotated data in a weak supervision setting, and building neural network-based frameworks to attack the problems of question answering.<\/p>\n\n\n\n<h3 id=\"hojoon-lee\">Hojoon Lee<\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/gsis.kaist.ac.kr\/cysec\/brentkang\/brentkang\" target=\"_blank\" rel=\"noopener noreferrer\">Brent Byunghoon Kang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>OS security, virtualization<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research goal is to enhance the security and reliability of modern systems from the operating system level. I am particularly interested in building secure computing architectures by using trusted execution environments (TEEs). Modern computing systems are under severe threat: software vulnerabilities are constantly discovered and malwares are infecting billions. TEEs provide a safe execution environment in which system integrity monitoring or system transactions can be performed. I am currently exploring various forms of secure kernel design that involves TEE. My goal is to make fundamental changes to the traditional kernel architectures such that it becomes more resilient against software attacks.<\/p>\n\n\n\n<h3 id=\"tae-hyun-oh\">Tae-Hyun Oh<\/h3>\n\n\n\n<p>Korea Advanced Institute of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/rcv.kaist.ac.kr\/\" target=\"_blank\" rel=\"noopener noreferrer\">In So Kweon<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision and machine learning (especially low-rank and sparse structure)<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interest is to investigate favorable non-convex models for tremendous applications. It has been known that global optimal solutions obtained from convex formulations may not often correspond to physically desirable solutions under many practical conditions. Since there is no limit of convexity while formulating non-convex models, flexible designs can be considered and the working range of algorithms can be broadened. In this stream, my future research will focus on: 1) Non-convex modeling that has favorable properties. It can be achieved by carefully looking into a problem and data. 2) Developing scalable optimizers corresponding to the newly developed formulations. I would like to make a stepping stone between engineering and theoretical people by providing useful and easy solutions.<\/p>\n\n\n\n<h3 id=\"yingwei-pan\">Yingwei Pan<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/staff.ustc.edu.cn\/~lihq\/index.html\" target=\"_blank\" rel=\"noopener noreferrer\">Houqiang Li<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Video understanding, large-scale visual search, click-through data analysis<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>One of the fundamental problems in vision and multimedia is to bridge the semantic gap between visual content and text. In our previous research, we studied the problem from the viewpoint of cross-view (visual-semantic) embedding, that is, integrate content, structure, and\/or click data for learning a joint space that connects different natures of vision and text. Following the research in this direction, I will further explore the complex semantic relationship between long video sequence and natural language by considering three main challenges: 1) learning an effective representation of a long video, 2) video understanding from individual tags to natural sentence, and 3) demonstrating the success of our technologies in a wide range of applications such as video temporal localization and sentence (story) generation.<\/p>\n\n\n\n<h3 id=\"hao-tang\">Hao Tang<\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/sei.pku.edu.cn\/~zhanglu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Lu Zhang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Software analysis and testing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Modern programming language features (e.g., callbacks, reflection, file I\/O) provide convenient language support for user-defined and platform-specific external behaviors. However, it is difficult to analyze unknown external behaviors with static analyzers. By now, we have made a small step towards analyzing these behaviors. I propose conditional reachability (formulated as tree-adjoining-language reachability) to handle callbacks when performing reachability analysis for software libraries. In my future work, I plan to improve the efficiency, scalability, precision, and generality of conditional reachability and apply it to real-life mobile and web applications. Also, I will continue to exploit new algorithms and data structures to analyze other types of unknown external behaviors.<\/p>\n\n\n\n<h3 id=\"tatsunori-taniai\">Tatsunori Taniai<\/h3>\n\n\n\n<p>The University of Tokyo<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.hci.iis.u-tokyo.ac.jp\/~ysato\/\" target=\"_blank\" rel=\"noopener noreferrer\">Yoichi Sato<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Mathematical optimization for low- and mid-level computer vision<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am interested in applications of mathematical optimization techniques for low- and mid-level computer vision problems, such as 3D shape recovery from images, dense correspondence estimation, and image segmentation. My long-term research goal is to establish a joint framework of image co-segmentation and dense correspondence estimation where common object regions between input images are precisely segmented and aligned as output. Such rich information about object regions and positions is useful in many applications such as fine-grained object detection, scene parsing, and image editing.<\/p>\n\n\n\n<h3 id=\"hao-wang\">Hao Wang<\/h3>\n\n\n\n<p>The Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.cse.ust.hk\/~dyyeung\/\" target=\"_blank\" rel=\"noopener noreferrer\">Dit-Yan Yeung<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Machine learning, data mining, Bayesian deep learning, social network analysis, recommender systems<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>There are two primary classes of tasks in AI. The first class is perception tasks like visual object recognition and text understanding. These are the tasks that can be easily completed by normal human beings. The second class is inference and planning. This involves a higher level of intelligence like decision making, data analysis, and logic deduction. In a sense, deep learning with its multiple processing layers is better at the first class of tasks and probabilistic graphical models with its Bayesian nature excel at the second class of tasks. The problem is that in the real world, these two classes often entangle with each other. Our research goal is to combine the power of deep learning and Bayesian models in a single principled framework to get the best of both worlds. We call this framework Bayesian Deep Learning. So far, we have successfully applied Bayesian Deep Learning for applications like recommendation, relational learning, and link prediction. In the future, we would like to extend our work both horizontally, to handle more applications, and vertically, to provide more theoretical guarantees along with more sophisticated models.<\/p>\n\n\n\n<h3 id=\"feng-long-xie\">Feng-Long Xie<\/h3>\n\n\n\n<p>Harbin Institute of Technology<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/homepage.hit.edu.cn\/pages\/lihaifeng\/\" target=\"_blank\" rel=\"noopener noreferrer\">Haifeng Li<\/a>, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/frankkps\/\" target=\"_blank\" rel=\"noopener noreferrer\">Frank K. Soong<\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Cross-lingual speech synthesis, voice conversion, speaker adaptation<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Cross-lingual TTS (text-to-speech) synthesis can be defined as synthesizing speech in the target language (L2) with a specific speaker\u2019s recorded speech in his native language (L1) and to maintain this speaker\u2019s voice characteristics. Currently, I am trying to use Speaker Independent Deep Neural Networks to equalize the speaker difference and Kullback-Leilbler divergence to measure the acoustic units\u2019 difference. In the future, I will apply this framework to many speech applications, e.g., voice conversion, speaker adaptation, speech recognition, language learning, and talking head.<\/p>\n\n\n\n<h3 id=\"ting-zhang\">Ting Zhang<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors: Yong Wang, <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/hon\/\" target=\"_blank\" rel=\"noopener noreferrer\">Hsiao-Wuen Hon<\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision, machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interests have revolved around computer vision and machine learning, with a specific focus on fast approximate nearest neighbor (ANN) search. The goal of fast ANN search is to find the closest item of the given query accurately as well the fastest of all of the database vectors. To achieve that goal, our previous work uses composite quantization to represent the database vectors by compact codes. My long-term research goal is to continue exploring algorithms to improve the search accuracy as well as the search efficiency of the approximate nearest neighbors. I believe that many applications\u2014such as similar image search, object retrieval, and so on\u2014can benefit from fast ANN search.<\/p>\n\n\n\n<h3 id=\"yongfeng-zhang\">Yongfeng Zhang<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.tsinghua.edu.cn\/publish\/csen\/4623\/2010\/20101224180015486255888\/20101224180015486255888_.html\" target=\"_blank\" rel=\"noopener noreferrer\">Shaoping Ma<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Personalization theories, computational economics, sentiment analysis<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I believe that academic research should advance itself not only for commercial benefits, but also for the benefit of society. Economists, philosophers, and sociologists devote their lives for a better human society in our physical world, and we as computer scientists should also push the virtual online society towards a more fair and just world. The research that I am targeting now and in the future falls into the emerging research field of computational economics. I especially care about the new frameworks for online services such as e-commerce, online financing, or freelancing based on the maximization of the joint surplus of both service consumers and providers (i.e., the social surplus) so that online services are matched from providers to consumers in a way that benefits the virtual society for a social good.<\/p>\n\n\n\n<h3 id=\"zhenzhe-zheng\">Zhenzhe Zheng<\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/cs.nju.edu.cn\/gchen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Guihai Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Algorithmic network economics, wireless network, mobile computing, and cloud computing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Generally, I am interested in research problems that lie at the interaction of economic and computer science. My recent research interests mainly focus on resource management in computer networks. With the rapid development of distributed systems, networking systems, and cloud computing technology, modern network resource management needs the cooperation of different network entities. However, the goals of selfish and rationale network entities usually conflict with the overall goal of the whole network system. My research goal is to apply the methodology of game theory to design efficient mechanisms for network resource management in different layers of network systems. I am also interested in designing efficient and economic-robust market systems, such as online ad auctions and data markets.<\/p>\n\n\n\n\n\n<h3 id=\"szu-wen-fan\">Szu-Wen Fan<\/h3>\n\n\n\n<p>Keio University<\/p>\n\n\n\n<p>Supervisor: Masahiko Inami<\/p>\n\n\n\n<p><b>Research interests: <\/b>Substitutional reality (SR), human computer interaction (HCI), cross-modal perception, human\u2019s perception of reality<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am enthusiastic about finding out how humans perceive reality, based on the multi-modal stimulation we receive with our five senses. As we are well aware, our beliefs about reality are limited to what we perceive with our sensory organs, which could be misdirected from true reality. Therefore, we can immerse users in alternate reality experiences by just stimulating authentic sensory feedbacks using a computer. However, these alternate reality experiences are limited to one reality only. Users are in either actual reality or alternate reality. I suggest that we can extend a human\u2019s perception of reality to be composed of more than one reality simultaneously, by focusing on cross-modal perception and seamless transition between actual and alternate reality. I am looking to blend realities:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>From remote reality, for immersive telepresence<\/li>\n\n\n\n<li>From past reality, for immersive memories<\/li>\n\n\n\n<li>From wearable augmented sensory reality, for augmenting human ability<\/li>\n<\/ol>\n\n\n\n<h3 id=\"seunghoon-hong\">Seunghoon Hong<\/h3>\n\n\n\n<p>Pohang University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/cv.postech.ac.kr\/~bhhan\" target=\"_blank\" rel=\"noopener noreferrer\">Bohyung Han<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision and machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am interested in developing a visual tracking system that can reliably track the target over \u201clong-term\u201d videos. Although substantial progress has been made in the tracking literature, this problem still remains challenging since most existing trackers are susceptible to model drift and temporary tracking failures. To resolve such issues, my previous research has mainly focused on developing reliable graphical models for tracking. By identifying a suitable order of tracking within a video, more reliable tracking can be achieved robust to intermediate failures. My future research would focus on long-term visual tracking based on high-level context understanding. Many real-world videos, such as movies and TV shows, contain drastic variations of the target and scene, and tracking based only on low-level visual information would inevitably fail with these challenges. I believe that contexture information, such as scene structure, object categories, and so forth, provides additional information to resolve ambiguities in tracking and to find more meaningful structure of the video for tracking.<\/p>\n\n\n\n<h3 id=\"wenping-hu\">Wenping Hu<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors: Yong Wang, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/hon\/\" target=\"_blank\" rel=\"noopener noreferrer\">Hsiao-Wuen Hon<\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer-aided language learning, speaker recognition, speech recognition, deep learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Computer-Aided Language Learning (CALL) systems, powered by the advancement of speech technology, can bridge the gap between the supply and demand of language teachers and have become ubiquitous learning tools with widely used mobile devices. We have built our own CALL system and transferred it into Microsoft Bing dictionary. This system is able to evaluate language learners\u2019 pronunciation quality, automatic detect the mispronunciation error within the word, and give some constructive feedbacks to language learners, e.g., common error patterns. Currently, I am using deep learning for pronunciation proficiency assessment and deficiency detection. We trained our golden pronunciation model by deep neural networks, investigated more robust pronunciation algorithms, applied transfer-learning for pronunciation deficiency detection and speeded up the real-time system to make it practical. In the future, we will further extend our algorithms for free-style, spontaneous speech assessment, which will have an assessment of learner\u2019s spoken pronunciation proficiency based on his or her free talk recordings, just like TOEFL oral test. Without given the read transcription, it will be more challenging, but more helpful and practical for second language learners.<\/p>\n\n\n\n<h3 id=\"yanyan-jiang\">Yanyan Jiang<\/h3>\n\n\n\n<p>Nanjing University<\/p>\n\n\n\n<p>Supervisor: Jian Lu<\/p>\n\n\n\n<p><b>Research interests: <\/b>Testing and analysis of concurrent programs<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Software quality assurance is becoming a major challenge for the modern software industry, as software products are getting larger, more complex, and parallel. Software testing practice, being the most prevalent approach for improving software quality, is also encountering emerging challenges. I am particularly interested in testing and analysis techniques to improve the quality of concurrent programs. Recently, researchers have proposed deterministic replay, symbolic analysis, and predictive analysis. These techniques points out a new trend for testing real-world concurrent software. I am conducting my research towards the following goals:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Record and replay a concurrent program execution that can last days long within affordable resource consumption<\/li>\n\n\n\n<li>Design sound and semantic-aware causal execution model for concurrent programs that can be checked in polynomial time<\/li>\n\n\n\n<li>Using symbolic analysis and constraint solving technique to systematically explore interleaving space of concurrent programs<\/li>\n<\/ol>\n\n\n\n<h3 id=\"hiroshi-kajino\">Hiroshi Kajino<\/h3>\n\n\n\n<p>The University of Tokyo<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.ibis.t.u-tokyo.ac.jp\/yamanishi\/index_e.html\" target=\"_blank\" rel=\"noopener noreferrer\">Kenji Yamanishi<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.geocities.co.jp\/Technopolis\/5893\/index_e.html\" target=\"_blank\" rel=\"noopener noreferrer\">Hisashi Kashima<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Human computation, crowdsourcing, machine learning, data mining<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research goal is to establish a secure and reliable infrastructure for human computation. Human computation is an emerging computer science technique in which machines utilize humans as computational resources to tackle computationally hard problems. Crowdsourcing is often used to allow machines to access human resources via the web. However, it has been often pointed out that crowdsourcing brings about a reliability issue that a result of computation is sometimes unreliable. My research focuses not only on the reliability issue, but also on privacy issues that arise, for example, when a task is submitted to crowdsourcing and when workers in crowdsourcing return computation results. By addressing the privacy issues, the applicability of human computation will become wider than ever.<\/p>\n\n\n\n<h3 id=\"yuan-lin\">Yuan Lin<\/h3>\n\n\n\n<p>Fudan University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/homepage.fudan.edu.cn\/zhangzz\/en\/\" target=\"_blank\" rel=\"noopener noreferrer\">Zhongzhi Zhang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Random walks in complex network<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Random walks are a fascinating research topic attracting intensive attention within the scientific community, not only for their intrinsic mathematical beauty but also for their important implications to various scientific issues, such as target searching, node ranking, computer vision, to name just a few. My research goal will focus on the following two tasks.<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Discovering the behavior of biased random walks in complex networks. Although unbiased random walks have been widely investigated, many dynamics should be described by biased random walks, which is still an outstanding issue. My work will both numerically and analytically research the influence of different biases on the crucial random walk indicators.<\/li>\n\n\n\n<li>Applying new random walk mechanisms in practical problems. Based on the obtained exhibitions of various biased random walks, my research will explore the applications of such bias in virus spreading, link prediction, community detection, and so on.<\/li>\n<\/ol>\n\n\n\n<h3 id=\"yutao-liu\">Yutao Liu<\/h3>\n\n\n\n<p>Shanghai Jiao Tong University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/ipads.se.sjtu.edu.cn\/haibo_chen\" target=\"_blank\" rel=\"noopener noreferrer\">Haibo Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Cloud and mobile security<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My primary research goal is to improve server and mobile security in system perspective. My current research projects investigate two areas of focus:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Leveraging potential architecture support to improve the dependability and security of current systems, specifically, using hardware features in different architectures to protect data privacy, enforce control flow integrity, actively monitor the whole system, timely detect potential attack, and so forth.<\/li>\n\n\n\n<li>Building and optimizing reliable virtualization environment to benefit system security. Virtualization has been a tremendous success, especially in server consolidation and security enhancement. However, there are still some problems when considering security aspects, and my goal is to ameliorate virtualization architecture to adapt to increasing security requirements.<\/li>\n<\/ol>\n\n\n\n<h3 id=\"canyi-lu\">Canyi Lu<\/h3>\n\n\n\n<p>National University of Singapore<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.ece.nus.edu.sg\/stfpage\/eleyans\/\" target=\"_blank\" rel=\"noopener noreferrer\">Shuicheng Yan<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cis.pku.edu.cn\/faculty\/vision\/zlin\/zlin.htm\" target=\"_blank\" rel=\"noopener noreferrer\">Zhouchen Lin<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Compressive sensing of block diagonal affinity matrix, convex optimization<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am interested in developing algorithms for important problems with support in theory. In particular, I am aiming to establish two or three representative works in my PhD career. They include:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Compressive sensing of block diagonal affinity matrix Affinity matrix is almost everywhere. The ideal affinity matrix should be block diagonal, i.e., data points from different subspaces or subjects are not similar at all. There are many known methods, such as sparse representation and low-rank representation, which lead to the block diagonal solution under certain conditions. We have established the Enforced Block Diagonal (EBD) conditions to unify many previous methods. We are aiming to find a convex model for block diagonal affinity matrix pursuit and to establish the theory for the exact recovery.<\/li>\n\n\n\n<li>Alternating Direction Method of Multiplier (ADMM) ADMM is the most important solver for convex optimization with linear constraint. It regains much attention in recent years due to the success of compressive sensing. There are many ADMM variants based on different structures of the objective function, but their convergence proofs are given case by case. We are aiming to give a general framework that unifies many known ADMMs with a unified proof. Also, there are still many open problems with ADMM. I aim to give a comprehensive study in the future.<\/li>\n<\/ol>\n\n\n\n<p>I am also interested in applying our algorithms and theory for computer vision and pattern recognition.<\/p>\n\n\n\n<h3 id=\"yongtae-park\">Yongtae Park<\/h3>\n\n\n\n<p>Korea University<\/p>\n\n\n\n<p>Supervisor: Hyogon Kim<\/p>\n\n\n\n<p><b>Research interests: <\/b>Software defined radio for mobile devices, vehicular networking<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am working on Software Defined Radio (SDR) for mobile devices. Recently, I demonstrated the feasibility of ZigBee communication on smartphones and am working to show the possibility of supporting various protocols without using dedicated chips. Providing these protocols as an app will catalyze the development and maintenance of protocols on future mobile devices. Also, it will enable people to develop the personalized and secure protocols for themselves. To make this happen, I am working on various issues such as optimization of SDR processing, fairness of antenna sharing, and so forth. Ultimately, my long-term research goal is to make mobile devices speak thousands of protocols. I hope my research plays an important role to facilitate \u201cInternet of Things\u201d by helping to connect humans with ambient smart devices.<\/p>\n\n\n\n<h3 id=\"shaoqing-ren\">Shaoqing Ren<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/staff.ustc.edu.cn\/~binli\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bin Li<\/a>, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/jiansun\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jian Sun<\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision, detection, and localization for face and generic objects<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interest lies in computer vision, especially detection and localization. Detection and localization as the fundamental problems in computer vision have been studied widely and are closing to practical applications. My previous research focuses on highly efficient detection and localization (alignment) on human face and generic objects, which makes these computer vision technologies cheaper on computation cost and closer to everybody\u2019s daily life. I envision more and more applications will benefit from the breakthrough of image understanding, especially apps on mobile devices. My future research will continue to explore accurate and highly efficient detection\/localization systems and make them more practical for real applications.<\/p>\n\n\n\n<h3 id=\"yingxia-shao\">Yingxia Shao<\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/net.pku.edu.cn\/~cuibin\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bin Cui<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Large-scale graph analysis, parallel computing framework, scalable data processing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>With the rapid growth of social networks, e-commence, and sensor networks, large-scale graphs become easily available in public. These graphs represent the complex relationships in the real world and contain a lot of valuable knowledge. Therefore, mining and analyzing these large-scale graphs is an important approach to discover the knowledge. The long-term goal of my research is to scale graph analysis to the scales of real-world graphs and develop a scalable and efficient large graph analysis toolkit. To achieve this goal, we design new graph algorithms according to several principles. First, the new algorithms will exploit the new characteristics of graphs, such as sparse, dynamic, and power-law distribution. Second, the workload patterns of graph problems will be unraveled as well. Lastly, for intricate graph metrics (e.g., SimRank, closeness, betweenness, and so forth), we will design new approximation algorithms. The toolkit will be carefully designed by abstracting primitive operations among above graph analysis task.<\/p>\n\n\n\n<h3 id=\"shuai-yi\">Shuai Yi<\/h3>\n\n\n\n<p>The Chinese University of Hong Kong<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.ee.cuhk.edu.hk\/~xgwang\/\" target=\"_blank\" rel=\"noopener noreferrer\">Xiaogang Wang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision and crowd video surveillance<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am now focusing on crowd video analysis in computer vision. Nowadays, with steady population growth and worldwide urbanization, crowded situations are more common. Crowd management and traffic control are common problems in public areas with a high population density. Millions of surveillance cameras are capturing these areas day and night. However, it is still very challenging to analyze these videos accurately and efficiently. A pedestrian\u2019s decision-making is very complex and many factors may have great influence on one\u2019s walking behavior. For example, the structure of the scene, the inner belief of the pedestrian, and the interaction with other moving pedestrians and stationary groups will all change pedestrians\u2019 walking paths and make the problem difficult to solve. I am now focusing my analysis on human behaviors in crowds and trying to find the relationship between a pedestrian\u2019s walking behavior and corresponding influence factors. In the future, I hope a joint pedestrian decision making model can be built which will aid in behavior analysis and abnormal detection.<\/p>\n\n\n\n\n\n<h3 id=\"chen-cao\">Chen Cao<\/h3>\n\n\n\n<p>Zhejiang University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.kunzhou.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kun Zhou<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer graphics, real-time facial animation<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research goal is to propose a real-time facial animation system for average users. Several issues need to be addressed to implement such a system. Primarily, the usage of special equipment such as markers or camera arrays should be avoided. Alternatively, an ordinary web camera should be the only capture device. Our system expectations are to be used in a wide range of environments, including both indoor and outdoor environments exposed to direct sunlight. The system should be capable of handling the changes in lighting and background. It is also vital to ensure maximum high performance system capabilities with the system. I will continue to focus on researching real-time facial animation in the future. There are many problems that need to be addressed and fixed with Real-time Facial Animation. Our current system requires a large amount of training data during the preprocessing step. It is ideal to find a solution to reduce the training data requirements, because the user needs to perform 60 different facial expressions and head poses. A key goal is to implement our real-time facial animation system on mobile devices. This, however, is not straightforward. Due to the limited computational power on mobile devices, the three-dimensional shape regression is time consuming. The limitations in our current system to handle dramatic lighting changes must be addressed since mobile devices are frequently used in various environments.<\/p>\n\n\n\n<h3 id=\"dong-chen\">Dong Chen<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/staff.ustc.edu.cn\/~binli\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bin Li<\/a>, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/jiansun\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jian Sun<\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision, face recognition, face detection<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Face recognition has been widely studied for more than three decades. Facial recognition of an image is divided into four steps: face detection, face alignment, feature extraction, and feature learning. Each element will have an impact on the overall performance. My long-term research goal is to improve face recognition accuracy during the four-step process. The following implementations will highlight my long-term research goal.<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>A more efficient structure for face detection and the face alignment pipeline will be designed. Face recognition\u2019s speed and accuracy, and the Landmark\u2019s Localization Algorithm will be improved.<\/li>\n\n\n\n<li>An illumination invariant feature combined with data-driven based approaches to increase face recognition accuracy under uncontrolled extreme lighting conditions will be designed.<\/li>\n\n\n\n<li>I will extend my feature and learning method to more face-related tasks such as attribute recognition, as well as exploiting these technologies to further promote the recognition rate.<\/li>\n<\/ol>\n\n\n\n<h3 id=\"luwei-cheng\">Luwei Cheng<\/h3>\n\n\n\n<p>The University of Hong Kong<\/p>\n\n\n\n<p>Supervisors: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cs.hku.hk\/~clwang\" target=\"_blank\" rel=\"noopener noreferrer\">Cho-Li Wang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cs.hku.hk\/~fcmlau\" target=\"_blank\" rel=\"noopener noreferrer\">Francis Chi Moon Lau<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Virtualization technologies for cloud computing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Virtualization is largely adopted in datacenters to support on-demand cloud computing. Current virtualization technologies are primarily invented to host unmodified or slightly modified operating systems using virtual machines (VMs). However, the existence of the hypervisor, an additional layer between the hardware and the guest operating system has significantly changed the assumptions of many protocols, which are originally designed for physical running environments. In particular, VM scheduling delays due to CPU sharing can negatively affect many kernel services such as the process scheduler, the disk scheduler and network protocols. Consequently, my research efforts go towards deeply integrating the hypervisor and the guest OS to form a more lightweight container for upper-layer applications.<\/p>\n\n\n\n<h3 id=\"quan-fang\">Quan Fang<\/h3>\n\n\n\n<p>Chinese Academy of Sciences<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/nlpr-web.ia.ac.cn\/mmc\/homepage\/csxu.html\" target=\"_blank\" rel=\"noopener noreferrer\">Changsheng Xu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Georeferenced social media research and application<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interest lies in georeferenced social media research and application. I focus on geo-social computing in particular, which aims to sense and mine the massive georeferenced social media data to understand the user and geo-location for better serving. I am working towards the following strategies. Georeferenced media mining and modeling aims to sense and mine the massive geo-tagged data to understand geo-location, especially from the human-sensed perspective. User modeling in social media is to understand users by exploiting rich online user-generated content. I will combine and model the user, geo-location, and content in a unified framework. I am dedicated to developing efficient and effective techniques for harvesting knowledge about users and geo-locations from georeferenced social media data. The derived knowledge can benefit various user-centric applications.<\/p>\n\n\n\n<h3 id=\"ping-luo\">Ping Luo<\/h3>\n\n\n\n<p>The Chinese University of Hong Kong<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.ie.cuhk.edu.hk\/people\/xotang.shtml\" target=\"_blank\" rel=\"noopener noreferrer\">Xiaoou Tang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision, computer graphics, and machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am interested in the areas of computer vision, computer graphics, and machine learning. My particular focus is deep learning and its applications with vision and graphics problems, including face and human detection, parsing (poses, clothing), attribute analysis, and identification. I envision that information technology in the future can help automatically find, identify, and connect people all over the world from various media\u2014such as photographs, social networking, and images on the Internet\u2014making everybody\u2019s life easier. My future research heads toward this goal both theoretically and practically. Image representation is a fundamental problem in computer vision. Although this representation can be directly learned from data by deep learning, how to learn or fuse the representations from different domains is still a great challenge. I will first devise a cross-domain multi-task method for image modeling so that it can fuse the data from different sources, while serving as the building block for many high-level tasks; such as detection, recognition, and segmentation. High computational cost is another challenge when working with large amounts of data. My second task will focus on building a large-scale distributed deep-learning framework, which can effectively deal with tens of millions of image data.<\/p>\n\n\n\n<h3 id=\"masayasu-ogata\">Masayasu Ogata<\/h3>\n\n\n\n<p>Keio University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" title=\"\" href=\"http:\/\/www.ayu.ics.keio.ac.jp\/~michita\/\" target=\"_blank\" rel=\"noopener noreferrer\">Michita Imai<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Human-computer interaction, natural user interface<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am interested in the area of HCI, especially natural user interface and the embodied interaction with sensing technique design and development, as well as the wearable device and interactive system. Our experiences with the living environment are getting complicated with user motivation and desired functionality. To ease and illuminate interfaces in daily life, I will propose a vision and an idea to increase interface transparency that makes the user unaware of the computer\u2019s existence when they access the everyday environment. Sensing techniques and designing appropriate feedback to the user are required for such an achievement. My experience and skill will reinforce the long-term goal that integrates the human and the system in a daily environment in the near future.<\/p>\n\n\n\n<h3 id=\"jianping-shi\">Jianping Shi<\/h3>\n\n\n\n<p>The Chinese University of Hong Kong<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cse.cuhk.edu.hk\/~leojia\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jiaya Jia<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Computer vision, machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I have an extensive interest in computer vision and machine learning. My previous research focuses mainly on the model\u2019s point of view, including Sparse and Low-rank Matrix Factorization. These models have been widely used in many applications. To be proficient in modeling is not enough. I believe building appropriate models to fit the vision data is much more important. I am currently working on several projects, including Image Blur Analysis, Human Detection, and RGBD Video Processing. Such projects necessitate a deep understanding of the image data. For example, the Blur Analysis is based on several natural image statistics. Our Human Co-detection System explores the intrinsic similarity for the same person across photos. The RGBD Visual Tracking utilizes the depth distribution of the tracking object. All these projects stem from vision data understanding. My future research will continue to explore computer vision\u2019s interesting aspects, while also advancing the computer to understand our colorful world.<\/p>\n\n\n\n<h3 id=\"shusen-wang\">Shusen Wang<\/h3>\n\n\n\n<p>Zhejiang University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/bcmi.sjtu.edu.cn\/~zhzhang\/\" target=\"_blank\" rel=\"noopener noreferrer\">Zhihua Zhang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Machine learning, matrix analysis, convex optimization<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research focus is on enabling large-scale machine learning using randomized approximations. As we know, many classical machine learning methods suffer from high time and space complexities and are thus prohibitive in big-data applications. Some randomized approximation approaches\u2014for example, random projection, random selection, and the Nystrom method\u2014reduce the time and space complexities of many machine-learning methods from quadric or cubic to near linear by sacrificing a little accuracy. I am working on improving the existing approximation methods, devising new approximation methods, and providing theoretical guarantees for the methods. My goal is to make the randomized approximation approaches more accurate and more efficiently computed.<\/p>\n\n\n\n<h3 id=\"huanjing-yue\">Huanjing Yue<\/h3>\n\n\n\n<p>Tianjin University<\/p>\n\n\n\n<p>Supervisors: Feng Wu, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.tju.edu.cn\/seie\/english\/fas\/faculty\/201106\/t20110605_105661.html\" target=\"_blank\" rel=\"noopener noreferrer\">Jingyu Yang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Cloud-based image coding and processing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My long-term research goal is to advance the state-of-the-art technologies in signal processing and computer vision through cloud data. The number of images shared on the Internet is dramatically increasing due to the dramatic increase in social websites. This provides a live and huge image base, which has posed demands for the development and implementation of content-based image retrieval. In addition, it brings new opportunities for many well-known, ill-posed image processing and computer vision problems, such as image composition, completion, and colorization. I mainly focus on cloud-based image coding, super-resolution, and denoising. Different from the traditional methods, our work proposes utilizing the external correlated images to improve coding efficiency and image restoration quality. Our work is an early step toward a new era of cloud-based image coding and processing.<\/p>\n\n\n\n<h3 id=\"jun-zhang\">Jun Zhang<\/h3>\n\n\n\n<p>Nanyang Technological University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www3.ntu.edu.sg\/home\/xkxiao\/\" target=\"_blank\" rel=\"noopener noreferrer\">Xiaokui Xiao<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests: <\/b>Database management and data privacy<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>There is a growing emphasis in the world at large on utilizing and disseminating aggregate statistics from demographic data, health records, Internet activity, and other sources. This data, however, can seldom be accessed for public studies due to concerns over individual privacy. Existing methods for privacy preservation still offers inadequate assurance, because they rely on restrictive assumptions on how a malicious user may attack the data, whereas those assumptions can be easily violated in practice. Motivated by the practical importance of privacy protection and the deficiencies of the existing methods, my research work aims to conduct a comprehensive study on mathematically rigorous privacy protection models. Specifically, I am interested in:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Generic and fundamental algorithms under Differential Privacy, a state-of-the-art privacy model that is rigorously formulated based on statistics, offering an extremely strong privacy guarantee;<\/li>\n\n\n\n<li>Novel class of privacy definitions with better trade-off between privacy and utility.<\/li>\n<\/ol>\n\n\n\n\n\n<h3 id=\"yang-cao\">Yang Cao<\/h3>\n\n\n\n<p>Beihang University<\/p>\n\n\n\n<p>Supervisor: Jinpeng Huai<\/p>\n\n\n\n<p><b>Research interests:<\/b> Database theory and systems, graph pattern matching, social searching<\/p>\n\n\n\n<p><b>Long-term research goal:<\/b><\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Graph pattern matching revised for social searching: Graph pattern matching is fundamental to social analysis. Traditional techniques like subgraph isomor-phism and graph simulation either impose too strong a topological constraint on graphs to retrieve meaningful matches and incur high computational complexity (for example, subgraph isomorphism is NP-complete), or are too loose to find correct matches. Moreover, data graphs for modeling social networks are usually evolving over time. Existing models and techniques often fall short of handling this without incurring high computational complexity. To this end, I will propose new graph pattern matching models (queries) as well as techniques (algorithms) for evaluating them. The model should be well-designed such that it could strike a balance between the express power (semantics) and the computational complexity for evaluating them, and should be able to handle the evolvement of data graphs as well.<\/li>\n\n\n\n<li>Keywords search on graph data: Keywords searching has been a longstanding issue for search engines. Traditionally, keyword searching is often conducted on structured and semistructured data, as well as documents. Existing models and techniques for keywords searching cannot be naively adopted for social search engines. Recent research about keywords searching on graph data mainly focuses on developing techniques for retrieving matched results on graph data. My research for this aspect will instead turn to proposing new keywords searching models specific for graph data, to take the advantages of graph pattern queries to provide more accurate semantics for emerging applications.<\/li>\n\n\n\n<li>New computational model and complexity classes: In the context of large-scale data, traditional computational classes are too generic for identifying problems that are solvable practically, and moreover, they cannot take properties of data graphs (social networks) into consideration. To this end, I will propose a computational model and a hierarchy of complexity classes, to provide a dichotomy between those queries that are feasible on large-scale data graphs and those that are not, and thus properly classify social searching problems, with regard to the express power and evaluating complexity of the queries.<\/li>\n<\/ol>\n\n\n\n<h3 id=\"menglei-chai\">Menglei Chai<\/h3>\n\n\n\n<p>Zhejiang University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/kunzhou.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kun Zhou<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Computer Graphics, Computer Vision, Image Processing, and so forth<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My long-term research goal primarily targets portrait image manipulation, especially for recovering semantic information from common images and providing a series of techniques which enables useful applications for average users. As we know, images and videos featuring humans as their subjects are of great interest for both industrial experts and average end users, and have motivated various kinds of research in computer graphics during the past decades. But few of them involve 3-D information to resolve the occlusions and ambiguities issues, making these low-level image processing methods of limited application for specific tasks of portrait manipulation. Our research work will focus on portrait image manipulation, especially for components such as face, hair, body shape, clothing, and so forth. The key thought is to extract 3-D information by utilizing specific prior knowledge achieved with machine-learning techniques, and apply them to drive a broad range of user-friendly applications which were previously challenging.<\/p>\n\n\n\n<h3 id=\"jun-kato\">Jun Kato<\/h3>\n\n\n\n<p>The University of Tokyo<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www-ui.is.s.u-tokyo.ac.jp\/~takeo\/\" target=\"_blank\" rel=\"noopener noreferrer\">Takeo Igarashi<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Human-computer interaction and programming language (user interface for programmers)<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>I am interested in the broad area of human-computer interaction, but have been especially focused on interactions between humans and the real world through computers (physical computing and robot applications), their input modalities (natural user interfaces), and their development methods (prototyping toolkits and integrated development environments). In my future vision, everyone can be a weekend carpenter who makes his\/her everyday life easy and comfortable with help of information technology\u2014in other words, end-user programming of the real world in the real world. Currently, I am planning to take two steps toward this goal.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>First, I will propose development environments that improve the programmer\u2019s experience (Programmer\u2019s eXperience, PX) on \u201creal-world programming\u201d that involves interactions with the real world, such as robot control and vision-based object recognition.<\/li>\n\n\n\n<li>Second, I will propose end-user development environments for the same \u201creal-world programming\u201d that make creation of customized applications feasible to end users through specialized user interfaces.<\/li>\n<\/ul>\n\n\n\n<p>Given my past experience on PX research, I expect that further investigation on the professional user interfaces will serve as the fundamentals for designing new user interfaces for the end users.<\/p>\n\n\n\n<h3 id=\"yin-hsi-kuo\">Yin-Hsi Kuo<\/h3>\n\n\n\n<p>National Taiwan University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.csie.ntu.edu.tw\/~winston\/\" target=\"_blank\" rel=\"noopener noreferrer\">Winston H. Hsu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Multimedia content analysis, image retrieval, and mobile visual search<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Nowadays, the rapid computational power of mobile devices brings the emerging need for mobile visual search. Different from the traditional content-based retrieval system, the mobile devices have the ability to process the query (for example, feature extraction) before the transmission. In recent years, the hash-based method has become promising for approximate nearest neighbor (ANN) search because of its ability to deal with high-dimensional features and large-scale databases in an efficient way. An efficient and effective method on the mobile devices is extremely crucial; hence, the hash-based approach becomes a possible direction in reality. In the future, we attempt to integrate the hash-based approach with contextual information to achieve efficient and scalable mobile visual search.<\/p>\n\n\n\n<h3 id=\"hongjin-liang\">Hongjin Liang<\/h3>\n\n\n\n<p>University of Science and Technology of China<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/staff.ustc.edu.cn\/~xyfeng\/\" target=\"_blank\" rel=\"noopener noreferrer\">Xinyu Feng<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Program verification<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>The computer industry has shown explosive growth in the past years. This trend is not likely to slow down. However, the lack of reliable and secure software is becoming a bottleneck for such growth. In particular, programming on multiprocessors is extremely error-prone, but very difficult to debug. Thus, in my long-term research career, I would like to develop new and easy-to-use technologies and tools that could improve the reliability, safety, and security of software on multiprocessors. I am particularly interested in:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Designing program logics for verifying the correctness (including safety and liveness properties) of concurrent programs<\/li>\n\n\n\n<li>Developing automatic or semi-automatic tools to support concurrent program verification<\/li>\n\n\n\n<li>Applying my technologies and tools to verify today\u2019s concurrent libraries (such as java.concurrent.util), runtime support in concurrent systems (such as concurrent garbage collectors), and other concurrent software used in practice<\/li>\n<\/ol>\n\n\n\n<h3 id=\"yongxin-tong\">Yongxin Tong<\/h3>\n\n\n\n<p>Hong Kong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/www.cs.ust.hk\/~leichen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Lei Chen<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Database and data mining, machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interests mainly focus on uncertain data management and mining. With the emergence of many real applications\u2014such as sensor network monitoring, moving object search, protein-protein interaction (PPI) network analysis, and so forth\u2014managing and mining uncertain data has attracted much attention in the database and data mining communities recently. Although current researches have solved some fundamental operations over uncertain data\u2014for example, join, ranking, mining frequent item sets, clustering, and so forth\u2014there are a few works to explore the hidden correlation in uncertain data due to the intricate probabilistic structure and high computational complexity. Therefore, my research aims to address these challenges via incorporating both theoretical and practical viewpoints.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>On the theoretical side, I will first design a novel model to capture intrinsic correlated properties in uncertain data. Additionally, I will propose a series of efficient and effective algorithms in order to adapt complex structural data, such as uncertain data streams and uncertain graphs. This is particularly important in the age of big data as well.<\/li>\n\n\n\n<li>On the practical side, I hope to extend our theoretical model and algorithms to real application scenarios, in other words, I try to develop a new crowdsourcing platform based on mobile computing and utilize our probabilistic model to handle the uncertainty control in this system, which is one of most urgent problems in the crowdsourcing researches.<\/li>\n<\/ul>\n\n\n\n<p>In summary, my research goal is to better discover and manage the hidden correlation and rules over massive uncertain data effectively.<\/p>\n\n\n\n<h3 id=\"xinggang-wang\">Xinggang Wang<\/h3>\n\n\n\n<p>Huazhong University of Science and Technology<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/mc.eistar.net\/MCWebDisplay\/PersonDetails.aspx?Name=Wenyu%20Liu\" target=\"_blank\" rel=\"noopener noreferrer\">Wenyu Liu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Computer vision and machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interests are computer vision and machine learning, especially the problem of object detection. Object detection concerning detecting instances of semantic objects of a certain class in digital images and videos, is a fundamental problem in computer vision, and has wide applications in people\u2019s lives, for example, face recognition, image search, automatic driving, and so forth. Through years of research, the problem of frontal face detection has been well studied, and this technology has been integrated into products. However, detecting most of the other objects in the real world is a very difficult problem due to the great variations of object appearances, such as persons and cars. My previous research mainly focuses on part-based object detection, shape-based object detection, and semi-supervised learning for object detection. It brings together shape feature design, object model design, machine learning, and optimization. Based on my previous research, my long-term research goal is to build an object detection system in a very weakly supervised way, which can achieve the state-of-the-art performance and be used in some real-life applications. I will put more effort on:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Studying discriminative object part detector; for example, people can easily recognize a leopard when looking at dapple of leopard. Part detector is robust to occlusion. Ensemble of part detector can reach good object detector.<\/li>\n\n\n\n<li>Using context information for object detection which combines scene understanding and object recognition.<\/li>\n\n\n\n<li>Combining modeling based methods with data-driven based methods.<\/li>\n<\/ol>\n\n\n\n<h3 id=\"yizhong-zhang\">Yizhong Zhang<\/h3>\n\n\n\n<p>Zhejiang University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/kunzhou.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kun Zhou<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Computer graphics, physically based simulation<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My research interest is in physically based simulation, especially simulation of complex natural phenomena for graphics. Physically based simulation is a powerful tool in both science and engineering. It can help us to predict unknown phenomena that may prevent the success of an experiment or the improvement of a product\u2019s design before manufacturing. In graphics, we need simulations to make animations of water, fire, cloth, and many other phenomena that contain high frequency details. We want the simulation to be fast while being reliable so that it will be easier for artists to control the animation. My research goal is to improve the performance of simulation, including propose reduced physical model, method of efficient and robust handling of geometry, and parallel computing. By using these methods, we can get high quality simulation results more quickly.<\/p>\n\n\n\n<h3 id=\"xin-zhao\">Xin Zhao<\/h3>\n\n\n\n<p>Peking University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/net.pku.edu.cn\/~lxm\/\" target=\"_blank\" rel=\"noopener noreferrer\">Li Xiaoming<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Social media analysis, web text mining, machine learning<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>My general research interests are within the area of online social networks(for example,Twitter and Facebook) analysis. Online social networks\u2014in particular, their rapid growth and development\u2014are continuously attracting users all around the world, which significantly changes the way that people live. Although various text mining methods have been shown effective to deal with traditional document collections, for example, scientific publications, very few of them are tested to achieve very robust and sound performance on real data sets of online social networks due to the fact we still do not fully understand the nature of online social networks, such as their underlying information structure, user behaviors, and connecting patterns. To overcome these difficulties, the goal of my research is to develop both principled methodologies and innovative applications for automatically analyzing and discovering knowledge from online social networks. Specifically, I will mainly focus on the aspect of content analysis of online social networks in the long term. More challenging than traditional content analysis, we have to first understand the underlying information patterns and uncover the generative process of such information before we can construct effective models. Previous research experiences will be helpful for me to identify and solve real-world problems that are valuable to common users in this direction. During the problem-solving process, I will try to construct formal methods with clear and intuitive motivations, borrow the ideas and techniques from multiple disciplines, and evaluate research results with large-scale real data.<\/p>\n\n\n\n<h3 id=\"chong-zu\">Chong Zu<\/h3>\n\n\n\n<p>Tsinghua University<\/p>\n\n\n\n<p>Supervisor: <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/itcs.tsinghua.edu.cn\/yao\" target=\"_blank\" rel=\"noopener noreferrer\">Chi-Chih Yao<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p><b>Research interests:<\/b> Quantum information processing<\/p>\n\n\n\n<p><b>Long-term research goal: <\/b>Quantum information and computation is an interdisciplinary subject of computer science and quantum physics. During recent decades, there have been a lot of important theoretical results in this field, including unconditional secure cryptography, true random number generation, exponential speedup algorithm for factoring, and so on. However, realizing quantum information processing and quantum computation is still not an easy task. In my research, I am mainly focusing on two different physical systems, photonic qubit and NV center in diamond, both of which are promising candidates for future quantum information processing and computation.<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Photonic qubit: In my recent research, we are working on a project to theoretically propose and experimentally build a brand-new, reliable, and practical quantum random number generator. By saying reliable, it means that we can always use some methods to certify that the random number we are producing is indeed from quantum power, rather than some classical device or lack knowledge of the device, and the randomness (characterized by min-entropy) can be bounded; while saying practical means that our random number generator can be efficient and easy to be realized. In the future, we plan to generate quantum state with more qubits, with which we can demonstrate interesting quantum information protocols, as well as realize large-scale quantum information processing.<\/li>\n\n\n\n<li>NV center in diamond: Recently, we have started to build NV center system. We want to use it to realize room-temperature quantum memory, as well as solid state quantum repeater for long-range quantum information. Our long-term goal is to build a hybrid quantum computation and information network with different quantum systems including trapped ion, matter qubit, and photonic qubit, which can be treated as a prototype for a future genuine quantum network.<\/li>\n<\/ol>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><b>Dongzhe Ma<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Peiran Ren<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Xiaohui Bei<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Xun Cao<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Quan Wang<\/b>, Peking University<\/li>\n\n\n\n<li><b>Xiang Song<\/b>, Fudan University<\/li>\n\n\n\n<li><b>Chenglin Li<\/b>, Shanghai Jiao Tong University<\/li>\n\n\n\n<li><b>Guo Li<\/b>, Zhejiang University<\/li>\n\n\n\n<li><b>Shengjun Huang<\/b>, Nanjing University<\/li>\n\n\n\n<li><b>Jing Yuan<\/b>, University of Science and Technology of China<\/li>\n\n\n\n<li><b>Cuiling Lan<\/b>, Xidian University<\/li>\n\n\n\n<li><b>Guangmin Wang<\/b>, Xi\u2019an Jiao Tong University<\/li>\n\n\n\n<li><b>Yang Cao<\/b>, University of Science and Technology of Huazhong<\/li>\n\n\n\n<li><b>Qiang Hao<\/b>, Tianjin University<\/li>\n\n\n\n<li><b>Zhen Liao<\/b>, Nankai University<\/li>\n\n\n\n<li><b>Lei Cui<\/b>, Harbin Institute Technology University<\/li>\n\n\n\n<li><b>Xiaoshuai Sun<\/b>, Harbin Institute Technology University<\/li>\n\n\n\n<li><b>Bolei Zhou<\/b>, The Chinese University of Hong Kong<\/li>\n\n\n\n<li><b>Lin Ma<\/b>, The Chinese University of Hong Kong<\/li>\n\n\n\n<li><b>Nobuyuki Umetani<\/b>, The University of Tokyo<\/li>\n\n\n\n<li><b>Yefeng Liu<\/b>, Waseda University<\/li>\n\n\n\n<li><b>Hyunson Seo<\/b>, Yonsei University<\/li>\n\n\n\n<li><b>Yohan Chon<\/b>, Yonsei University<\/li>\n\n\n\n<li><b>Jaesik Park<\/b>, Korea Advanced Institute of Science and Technology<\/li>\n\n\n\n<li><b>Sangwon Seo<\/b>, Korea Advanced Institute of Science and Technology<\/li>\n\n\n\n<li><b>Kazi Rubaiat Habib<\/b>, National University of Singapore<\/li>\n\n\n\n<li><b>Gang Yu<\/b>, Nanyang Technological University<\/li>\n\n\n\n<li><b>Nikolay Gravin<\/b>, Nanyang Technological University<\/li>\n\n\n\n<li><b>Shuo-Hung Chen<\/b>, National Tsing Hua University<\/li>\n<\/ul>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><b>Haris Javaid<\/b>, The University of New South Wales<\/li>\n\n\n\n<li><b>Yue Deng<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Chongyang Ma<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Xiong Li<\/b>, Shanghai Jiao Tong University<\/li>\n\n\n\n<li><b>Bo Geng<\/b>, Peking University<\/li>\n\n\n\n<li><b>Shiliang Zhang<\/b>, Institute of Computing Technology, The Chinese Academy of Science<\/li>\n\n\n\n<li><b>Xiulian Peng<\/b>, University of Science and Technology of China<\/li>\n\n\n\n<li><b>Xiao Zhang<\/b>, Tsinghua University<\/li>\n\n\n\n<li><b>Xinying Song<\/b>, Harbin Institute Technology University<\/li>\n\n\n\n<li><b>Jinbao Wang<\/b>, Harbin Institute Technology University<\/li>\n\n\n\n<li><b>Wei Wu<\/b>, Peking University<\/li>\n\n\n\n<li><b>Linghe Kong<\/b>, Shanghai Jiao Tong University<\/li>\n\n\n\n<li><b>Liang Wang<\/b>, Nanjing University<\/li>\n\n\n\n<li><b>Ping Chen<\/b>, Nanjing University<\/li>\n\n\n\n<li><b>Weiwei Wu<\/b>, University of Science and Technology of China<\/li>\n\n\n\n<li><b>Xiaojun Qian<\/b>, The Chinese University of Hong Kong<\/li>\n\n\n\n<li><b>Dongxiao Yu<\/b>, The University of Hong Kong<\/li>\n\n\n\n<li><b>Lu Wang<\/b>, The Hong Kong University of Science & Technology<\/li>\n\n\n\n<li><b>Seokhwan Kim<\/b>, University of Tsukuba<\/li>\n\n\n\n<li><b>Adiyan Mujibiya<\/b>, The University of Tokyo<\/li>\n\n\n\n<li><b>Seungjin Lee<\/b>, Korea Advanced Institute of Science and Technology<\/li>\n\n\n\n<li><b>Tae-Joon Kim<\/b>, Korea Advanced Institute of Science and Technology<\/li>\n\n\n\n<li><b>Gae-won You<\/b>, Pohang University of Science and Technology<\/li>\n\n\n\n<li><b>Sungjin Lee<\/b>, Seoul National University<\/li>\n\n\n\n<li><b>Shenghua Gao<\/b>, Nanyang Technological University<\/li>\n\n\n\n<li><b>Yi-Ling Hsieh<\/b>, National Cheng Kung University<\/li>\n\n\n\n<li><b>Cheng-Te Li<\/b>, National Taiwan University<\/li>\n\n\n\n<li><b>Tsung-Te Lai<\/b>, National Taiwan University<\/li>\n<\/ul>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><b>Novi Quadrianto<\/b>,<b> <\/b>The Australian National University<\/li>\n\n\n\n<li><b>Ashnil Kumar<\/b>,<b> <\/b>The University of Sydney<\/li>\n\n\n\n<li><b>William Voorsluys<\/b>,<b> <\/b>The University of Melbourne<\/li>\n\n\n\n<li><b>Shenghua Liu<\/b>,<b> <\/b>Tsinghua University<\/li>\n\n\n\n<li><b>Hao Wen<\/b>,<b> <\/b>Tsinghua University<\/li>\n\n\n\n<li><b>Zhiwei Xiong<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\n\n\n\n<li><b>Dong Liu<\/b>,<b> <\/b>Harbin Institute of Technology<\/li>\n\n\n\n<li><b>Bo Yu<\/b>,<b> <\/b>Harbin Institute of Technology<\/li>\n\n\n\n<li><b>Litian Tao<\/b>,<b> <\/b>Beihang University<\/li>\n\n\n\n<li><b>Yufeng Li<\/b>,<b> <\/b>Nanjing University<\/li>\n\n\n\n<li><b>Wei Wang<\/b>,<b> <\/b>Nanjing University<\/li>\n\n\n\n<li><b>Huanhuan Cao<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\n\n\n\n<li><b>Xiaoyin Wang<\/b>,<b> <\/b>Peking University<\/li>\n\n\n\n<li><b>Jun Lang<\/b>,<b> <\/b>Beijing Institute of Technology<\/li>\n\n\n\n<li><b>Derek Hao Hu<\/b>,<b> <\/b>The Hong Kong University of Science & Technology<\/li>\n\n\n\n<li><b>Kaiming He<\/b>,<b> <\/b>The Chinese University of Hong Kong<\/li>\n\n\n\n<li><b>Tasuku Oonishi<\/b>, Tokyo Institute of Technology<\/li>\n\n\n\n<li><b>Yoshida Yuichi<\/b>, Kyoto University<\/li>\n\n\n\n<li><b>Jun Hatori<\/b>, The University of Tokyo<\/li>\n\n\n\n<li><b>Jongwuk Lee<\/b>, Pohang University of Science and Technology<\/li>\n\n\n\n<li><b>Jung-Tae Lee<\/b>,<b> <\/b>Korea University<\/li>\n\n\n\n<li><b>Bingjun Zhang<\/b>,<b> <\/b>National University of Singapore<\/li>\n\n\n\n<li><b>Lixin Duan<\/b>,<b> <\/b>Nanyang Technological University<\/li>\n\n\n\n<li><b>Yu-Chen Sun<\/b>,<b> <\/b>National Chiao Tung University<\/li>\n\n\n\n<li><b>Kai-yin Chen<\/b>,<b> <\/b>National Taiwan University<\/li>\n<\/ul>\n\n\n\n\n\n<ul class=\"wp-block-list\">\n<li><b>Feng Zhang<\/b>,<b> <\/b>Beijing Institute of Technology<\/li>\n\n\n\n<li><b>Xiangyi Meng<\/b>,<b> <\/b>Beijing Institute of Technology<\/li>\n\n\n\n<li><b>Tong Wu<\/b>,<b> <\/b>Beijing University of Post and Telecommunication<\/li>\n\n\n\n<li><b>Li Xu<\/b>,<b> <\/b>Chinese University of Hong Kong<\/li>\n\n\n\n<li><b>Jun Lang<\/b>,<b> <\/b>Harbin Institute of Technology<\/li>\n\n\n\n<li><b>Qingqing Zhang<\/b>,<b> <\/b>Institute of Acoustics, Chinese Academy of Sciences<\/li>\n\n\n\n<li><b>Yanyan Lan<\/b>,<b> <\/b>Institute of Applied Mathematics, Academy of Mathematics and Systems Science, Chinese Academy of Sciences<\/li>\n\n\n\n<li><b>Xiaoqin Zhang<\/b>,<b> <\/b>Institute of Automation, Chinese Academy of Sciences<\/li>\n\n\n\n<li><b>Xiubo Geng<\/b>,<b> <\/b>Institute of Computing Technology, Chinese Academy of Sciences<\/li>\n\n\n\n<li><b>Souneil Park<\/b>,<b> <\/b>Korea Advanced Institute of Science and Technology<\/li>\n\n\n\n<li><b>Ki-woong Park<\/b>,<b> <\/b>Korea Advanced Institute of Science and Technology<\/li>\n\n\n\n<li><b>Dafan Dong<\/b>,<b> <\/b>Nankai University<\/li>\n\n\n\n<li><b>Yi Huang<\/b>,<b> <\/b>Nanyang Technological University<\/li>\n\n\n\n<li><b>Yu-Lin Wang<\/b>,<b> <\/b>National Cheng Kung University<\/li>\n\n\n\n<li><b>Yi-Hsuan Yang<\/b>,<b> <\/b>National Taiwan University<\/li>\n\n\n\n<li><b>Yantao Zheng<\/b>,<b> <\/b>National University of Singapore<\/li>\n\n\n\n<li><b>Lijiang Chen<\/b>,<b> <\/b>Peking University<\/li>\n\n\n\n<li><b>Sunghyun Cho<\/b>,<b> <\/b>Pohang University of Science and Technology<\/li>\n\n\n\n<li><b>YongDeok Kim<\/b>,<b> <\/b>Pohang University of Science and Technology<\/li>\n\n\n\n<li><b>Dikan Xing<\/b>,<b> <\/b>Shanghai Jiao Tong University<\/li>\n\n\n\n<li><b>Jingjing Fu<\/b>,<b> <\/b>The Hong Kong University of Science and Technology<\/li>\n\n\n\n<li><b>Jun Hong<\/b>,<b> <\/b>The University of Hong Kong<\/li>\n\n\n\n<li><b>Florian Mueller<\/b>,<b> <\/b>The University of Melbourne<\/li>\n\n\n\n<li><b>Tomoaki Higo<\/b>,<b> <\/b>The University of Tokyo<\/li>\n\n\n\n<li><b>Pinyan Lu<\/b>,<b> <\/b>Tsinghua University<\/li>\n\n\n\n<li><b>Jialin Zhang<\/b>,<b> <\/b>Tsinghua University<\/li>\n\n\n\n<li><b>Qiming Hou<\/b>,<b> <\/b>Tsinghua University<\/li>\n\n\n\n<li><b>Kun Xu<\/b>,<b> <\/b>Tsinghua University<\/li>\n\n\n\n<li><b>SYifei Don<\/b>,<b> <\/b>University of New South Wales<\/li>\n\n\n\n<li><b>Hao Xu<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\n\n\n\n<li><b>Yuan Liu<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\n\n\n\n<li><b>Myung \u2013 Suk Song<\/b>,<b> <\/b>Yonsei University<\/li>\n<\/ul>\n\n\n\n\n\n\n\n<h2 id=\"the-fellowship-experience-in-their-own-words\">The fellowship experience: in their own words<\/h2>\n\n\n\n<p>Fellows talk about their experiences before, during, and after their fellowship tenure.<\/p>\n\n\n\n<div class=\"wp-block-buttons is-layout-flex wp-block-buttons-is-layout-flex\">\n<div class=\"wp-block-button\"><a data-bi-type=\"button\" class=\"wp-block-button__link wp-element-button\" href=\"https:\/\/youtu.be\/NdVH36ONjDA\">Watch the video <\/a><\/div>\n<\/div>\n\n\n\n<hr class=\"wp-block-separator has-css-opacity is-style-dots\" \/>\n\n\n\n<h3 id=\"opened-my-mind-about-research-and-careers\">Opened my mind about research and careers<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-style-default is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cI am very fortunate to have a relationship with Microsoft Research Asia, which I\u2019ve always admired as one of the best research laboratories in the world. I jumped into MSRA as an intern during my master\u2019s studies. Every day\u2019s discussions with my mentor were insightful, and I have developed a sense of what makes \u2018good\u2019 research and built courage to pursue challenging research in PhD. The Fellowship award, which back then looked too competitive for me, has given me confidence and opened my mind to many career possibilities.\u201d \u2014 <strong>Tatsunori Taniai<\/strong>, University of Tokyo, 2015 Fellow<\/p>\n<\/blockquote>\n\n\n\n<h3 id=\"alumni-community-for-academic-exchange\">Alumni community for academic exchange<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cDuring the application, both my PhD advisor and the head of department were very supportive and encouraging about my fellowship. The award aspects of the Microsoft Research Asia Fellowship that attract me most are the opportunity to do an internship in Microsoft Research Asia and to join an alumni community for academic exchange.\u201d \u2014 <strong>Yingwei Pan<\/strong>, University of Science and Technology of China, 2015 Fellow<\/p>\n<\/blockquote>\n\n\n\n<h3 id=\"develop-confidence-as-a-researcher\">Develop confidence as a researcher<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cThe Microsoft Research Asia Fellowship award has helped me develop confidence as a researcher and provides me encouragement to do high-quality research work as well as responsibility. The award gives me freedom to pursue the research topics that I am really interested in. It allows to get many opportunities to present and discuss with diverse and excellent researchers.\u201d \u2014<strong>Tae-Hyun Oh<\/strong>, Korea Advanced Institute of Science and Technology, 2015 Fellow<\/p>\n<\/blockquote>\n\n\n\n<h3 id=\"strong-motivation-for-my-research\">Strong motivation for my research<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cIt is a great pleasure to receive a Microsoft Research Asia Fellowship. This prize has given me strong motivation for my research. My research is about multimedia retrieval, and I know that Microsoft Research Asia is also focusing on the same area.\u201d \u2014 <strong>Yin-Hsi Kuo<\/strong>, National Taiwan University, 2012 Fellow<\/p>\n<\/blockquote>\n\n\n\n<h3 id=\"relentlessly-pursue-higher-learning\">Relentlessly pursue higher learning<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cI am especially grateful for the internship opportunity that comes with the Fellowship award. My internship period has been very exciting and I have gotten to work with many excellent people. They not only taught me computer-related knowledge and skills, but also team spirit, problem-solving approaches, and how to relentlessly pursue higher learning. It has also been eye-opening to talk to fellow research students with different research interests.\u201d \u2014 <strong>Canyi Lu<\/strong>, National University of Singapore, 2014 Fellow<\/p>\n<\/blockquote>\n\n\n\n<h3 id=\"collaborate-with-some-of-the-best-researchers-in-the-world\">Collaborate with some of the best researchers in the world<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cI think the Microsoft Research Asia Fellowship is a fantastic opportunity for young researchers like me to gain hands-on research experience in a large corporation. I get to collaborate with some of the best researchers in the world and the experience has helped me a lot in determining my own research direction. At the same time, the generous financial support provided by the Fellowship makes my PhD study a lot easier and my life more trouble-free.\u201d \u2014 <strong>Wei Bai<\/strong>, Hong Kong University of Science and Technology, 2015 Fellow<\/p>\n<\/blockquote>\n\n\n","protected":false},"featured_media":786595,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"msr-opportunity-type":[155533],"msr-region":[197903],"msr-locale":[268875],"msr-program-audience":[243724],"msr-post-option":[],"msr-impact-theme":[],"class_list":["post-215593","msr-academic-program","type-msr-academic-program","status-publish","has-post-thumbnail","hentry","msr-opportunity-type-grants-and-fellowships","msr-region-asia-pacific","msr-locale-en_us","msr-program-audience-students"],"msr_description":"The Microsoft Research Asia Fellowship Program aims to empower and encourage PhD students in the Asia-Pacific region to realize their potential in computer science-related research.","msr_social_media":[],"related-researchers":[],"tab-content":[{"id":0,"name":"About","content":"The Microsoft Research Asia Fellowship Program identifies the next generation of research leaders through a unique program that offers a combination of mentorship, research, networking, and academic opportunities to promising young candidates. Since its inception in 1999, the program has attracted over one thousand PhD candidates from top universities in the Asia-Pacific region, and 441 outstanding PhD students have been awarded the Microsoft Research Asia Fellowship honor. Fellows have gone on\u202fto become prominent researchers and influential individuals in academia and industry.\r\n\r\nFellows earn more than a simple scholarship; they enjoy\u202flong-term close engagement with Microsoft Research Asia, one of the leading computer science research labs in the world.\u202fEach winner has the opportunity to complete an internship, during which they participate in hands-on, advanced research at <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-asia\/\">Microsoft Research Asia<\/a> in Beijing, China.\r\n<h2>2021 Award schedule<\/h2>\r\n<ul>\r\n \t<li>June 30, 2021 - Application deadline<\/li>\r\n \t<li>Early September - Onsite interview (only for those candidates who pass through the first round of reviews)<\/li>\r\n \t<li>Early October - 2021 Fellows announced<\/li>\r\n<\/ul>\r\n<h2>Provisions of the award<\/h2>\r\nThe Microsoft Research Asia Fellowship Program offers some of the industry\u2019s most competitive incentives and rewards, including:\r\n<ul>\r\n \t<li>Cash awards of $10,000 USD, which can be used to support research initiatives and academic endeavors including, but not limited to: attending international research conferences\/events, and subsidizing paper and patent application fees.<\/li>\r\n \t<li>The guarantee of an optional three-month internship at Microsoft Research Asia.<\/li>\r\n \t<li>Opportunity for intensive and individualized mentorship programs with a dedicated senior researcher, to help facilitate the fellow\u2019s research until graduation.<\/li>\r\n \t<li>Opportunity to attend the Microsoft Research Asia Academic Conference.<\/li>\r\n \t<li>At the discretion of Microsoft Research Asia, advisors of fellows might be invited to participate in collaborative research projects and academic exchange events.<\/li>\r\n<\/ul>\r\n<h2>Eligibility criteria<\/h2>\r\n<ul>\r\n \t<li>Applicants must attend a university in mainland China, Hong Kong, Japan, Korea, Singapore, or Taiwan, and do research work that relates to computing topics in which Microsoft Research is concerned with (click on Research Areas at the top of the page for a full list).<\/li>\r\n \t<li>According to the definition of the normative period of PhD programs of the applicant\u2019s university and region, the applicant must graduate in 2023 or later. The applicants and their advisors will be asked for supporting documents, if needed.<\/li>\r\n \t<li>The Microsoft Research Asia Fellowship Committee has final determination of the interpretation of the above clauses.<\/li>\r\n<\/ul>\r\n<h2>Supplementary information<\/h2>\r\nMicrosoft has always been committed to the mission of empowering every person and every organization on the planet to achieve more and has actively promoted the extensive application of computer technology in various fields and in public welfare undertakings. With the progress of data acquisition, data processing, and artificial intelligence in recent years, computer technology is playing an increasingly important role in the research of basic science. The current problems and challenges faced by human society also require the urgent intersection and integration of basic scientific research and computer science. Based on the current trends and demand, the 2021 MSRA Fellowship will continue to welcome applicants in the field of <a href=\"https:\/\/en.wikipedia.org\/wiki\/Computational_science\" target=\"_blank\" rel=\"noopener\">Computational Science<\/a> in order to discover academic talents who are innovating computer theory and technology to promote the development of basic scientific research, and to support and encourage them to better carry out their potential research work.\r\n<h3>Related Computational Science subject areas<\/h3>\r\nPlease refer to the Research Areas &gt; Other Sciences menu in the top navigation of this site to learn more about these research areas:\r\n<ul>\r\n \t<li>Computational Ecology &amp; Environmental Studies\r\n<ul>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/making-air-quality-data-meaningful-coupling-objective-measurement-with-subjective-experience-through-narration\/\">Making Air Quality Data Meaningful: Coupling Objective Measurement with Subjective Experience through Narration<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/contract-design-for-afforestation-programs\/\">Contract Design for Afforestation Programs<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/architecting-datacenters-for-sustainability-greener-data-storage-using-synthetic-dna\/\">Architecting Datacenters for Sustainability: Greener Data Storage using Synthetic DNA<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/multispecies-bioacoustic-classification-using-transfer-learning-of-deep-convolutional-neural-networks-with-pseudo-labeling\/\">Multispecies Bioacoustic Classification Using Transfer Learning of Deep Convolutional Neural Networks with Pseudo-labeling<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/stay-ahead-of-poachers-illegal-wildlife-poaching-prediction-and-patrol-planning-under-uncertainty-with-field-test-evaluations\/\">Stay Ahead of Poachers: Illegal Wildlife Poaching Prediction and Patrol Planning Under Uncertainty with Field Test Evaluations<\/a><\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Computational Biomedical Science &amp; Public Health\r\n<ul>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/collaboration\/studies-in-pandemic-preparedness\/\">Studies in Pandemic Preparedness<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/an-empirical-comparison-of-preservation-methods-for-synthetic-dna-data-storage\/\">An Empirical Comparison of Preservation Methods for Synthetic DNA Data Storage<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/towards-compliant-data-management-systems-for-healthcare-ml\/\">Towards Compliant Data Management Systems for Healthcare ML<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/disentangling-human-error-from-the-ground-truth-in-segmentation-of-medical-images\/\">Disentangling Human Error from the Ground Truth in Segmentation of Medical Images<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/understanding-the-human-brain-using-brain-organoids-and-a-structure-function-theory\/\">Understanding the Human Brain using Brain Organoids and a Structure-Function Theory<\/a><\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Computational Physics\r\n<ul>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/solving-quantum-many-body-problem-artificial-neural-networks\/\">Solving the Quantum Many-Body Problem with Artificial Neural Networks<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.nature.com\/articles\/s41557-020-0544-y\" target=\"_blank\" rel=\"noopener\">Deep-neural-network solution of the electronic Schr\u00f6dinger equation<\/a> | Nature Chemistry<\/li>\r\n \t<li><a href=\"https:\/\/arxiv.org\/abs\/2005.00223\" target=\"_blank\" rel=\"noopener\">Pushing the limit of molecular dynamics with ab initio accuracy to 100 million atoms with machine learning<\/a><\/li>\r\n \t<li><a href=\"https:\/\/arxiv.org\/abs\/2010.08895\" target=\"_blank\" rel=\"noopener\">Fourier Neural Operator for Parametric Partial Differential Equations<\/a><\/li>\r\n \t<li><a href=\"https:\/\/arxiv.org\/abs\/2102.01010\" target=\"_blank\" rel=\"noopener\">Machine learning accelerated computational fluid dynamics<\/a><\/li>\r\n \t<li><a href=\"https:\/\/arxiv.org\/abs\/2010.03409\" target=\"_blank\" rel=\"noopener\">Learning Mesh-Based Simulation with Graph Networks<\/a><\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Computational Materials Science\r\n<ul>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/learning-composable-energy-surrogates-for-pde-order-reduction\/\">Learning Composable Energy Surrogates for PDE Order Reduction<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/how-machine-learning-can-help-select-capping-layers-to-suppress-perovskite-degradation\/\">How machine learning can help select capping layers to suppress perovskite degradation<\/a> | Nature Communication<\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/electrical-probes-of-the-non-abelian-spin-liquid-in-kitaev-materials\/\">Electrical Probes of the Non-Abelian Spin Liquid in Kitaev Materials<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/optimized-micromagnet-geometries-for-majorana-zero-modes-in-low-g-factor-materials\/\">Optimized micromagnet geometries for Majorana zero modes in low g-factor materials<\/a><\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Other computational science fields<\/li>\r\n<\/ul>\r\n<h3>Evaluation criteria of computational science area applicants<\/h3>\r\nIn the Computational Science area, The MSRA Fellowship program focuses on doctoral students who have promoted and expanded research work in the above fields through innovations in computer theory, algorithms, and systems and have made influential achievements. We also look at doctoral students who have made contributions to the development of computer technology by introducing the research methods, theories, and technologies of basic science into the computer science field."},{"id":1,"name":"How to Apply","content":"<h2>How to apply<\/h2>\r\n<strong>The 2021 Microsoft Research Asia Fellowship application period is now closed.<\/strong>\r\n<ul>\r\n \t<li>PhD students must apply directly for the scholarship.<\/li>\r\n \t<li>Applications must include:\r\n<ul>\r\n \t<li>An application form<strong>\u00a0<\/strong>(<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/04\/MSRA_Fellowship_Application_Form-2021.docx\" target=\"_blank\" rel=\"noopener noreferrer\">download template here<\/a>)<\/li>\r\n \t<li>Curriculum vitae<\/li>\r\n \t<li>Research Statement. This statement (of approximately 1,000 words) provides a summary of your research accomplishments and current work, and highlight your most outstanding contribution to the academia. It must discuss the future direction and potential of your work, especially your research plan for the next few years of PhD study.<\/li>\r\n \t<li>At least two letters of reference from established researchers familiar with your research. Of these, one letter should come from your advisor. Scanned copies of the signed recommendation letters should be sent, by those recommending you, directly to <a href=\"mailto:fellowra@microsoft.com\">fellowra@microsoft.com<\/a>.<\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Applications must be submitted via the <a href=\"http:\/\/cmtint.research.microsoft.com\/MSRAFellowship2021\" target=\"_blank\" rel=\"noopener\">online application tool<\/a> in any of the following formats: Word document, text-only file, PDF or ZIP. Email or hard copy applications will not be considered.<\/li>\r\n \t<li>Applications submitted to Microsoft will not be returned. Microsoft cannot assume responsibility for the confidentiality of information in submitted applications. Therefore, applications should not contain information that is confidential, restricted, or sensitive.<\/li>\r\n \t<li>Incomplete applications will not be considered.<\/li>\r\n \t<li>Due to the volume of submissions, Microsoft Research Asia cannot provide individual feedback on applications that do not receive fellowship awards.<\/li>\r\n<\/ul>"},{"id":2,"name":"FAQ","content":"<h2>Frequently asked questions<\/h2>\r\n[accordion]\r\n[panel header=\"What should be included in a research statement?\"] The research statement should provide a summary of your research accomplishments and current work and highlight your most outstanding contribution to the academia. Importantly, it should clearly discuss the future direction and potential of your work, especially your research plan for the next few years of PhD study. It should be technical but remain intelligible to any reviewer of the track (we have four general tracks, which you can find in the online application tool). Because your research statement may be read by a reviewer with same track but outside of your subdiscipline, it is important to keep the \u201cbig picture\u201d in mind. A strong research statement presents a readable, compelling, and realistic research agenda with clear potential results. Research statements can be weakened by overly ambitious proposals, a lack of clear direction, and a lack of big-picture focus.\r\n\r\n<strong>Some general advice:<\/strong>\r\n<ul>\r\n \t<li>The goal of the research statement is to introduce yourself to a search committee and make them excited about your research. This committee will likely contain scientists both in and outside of your subdiscipline. The statement may be around 1,000 words, keeping in mind that <em>you want people to read it and buy in, <\/em>so please don\u2019t make it too long or too short<em>. <\/em><\/li>\r\n \t<li>Think of the overarching theme guiding your main research subject area. Write an essay that lays out:\r\n<ul>\r\n \t<li>The main theme(s), why it is important, and what specific skills you will use to attack the problem.<\/li>\r\n \t<li>A couple of specific examples of problems you have already worked on with success \u2014\u00a0to build credibility and give people an idea of what it is you do.<\/li>\r\n \t<li>A discussion of the future direction of your research. This section should build on the above and be exciting to people both in and outside your field. If you think that your research could lead to answers for big, exciting questions \u2014\u00a0say so and reach for the stars.<\/li>\r\n \t<li>Tie it all off with a final paragraph that leaves the reader with a good overall impression of your research.<\/li>\r\n<\/ul>\r\n<\/li>\r\n<\/ul>\r\n<ul>\r\n \t<li>There is no need to mention all of your projects; stick to those that align with your overarching theme. Introducing no more than three (3) research works is strongly recommended.<\/li>\r\n \t<li>Pay attention to jargon. You want reviewers to understand everything in your statement. Remember that the goal is to make the search committee excited about you \u2014\u00a0and they won\u2019t be excited about something they can\u2019t understand.<\/li>\r\n \t<li>If you have something that sets you apart (e.g. a prestigious achievement in your field), you may want to include it.<\/li>\r\n \t<li>There are no excuses for spelling errors.<\/li>\r\n<\/ul>\r\n[\/panel]\r\n\r\n[panel header=\"Who should submit the applications?\"] Required application materials can be submitted directly online by the applicant. The online application tool is available at: <a href=\"https:\/\/cmtint.research.microsoft.com\/MSRAFellowship2021\" target=\"_blank\" rel=\"noopener\">https:\/\/cmtint.research.microsoft.com\/MSRAFellowship2021<\/a> [\/panel]\r\n\r\n[panel header=\"In what language and format must the application materials be?\"] All application materials must be in English. All documents submitted must be in Word document, text-only file, PDF, or ZIP file format. [\/panel]\r\n\r\n[panel header=\"Students from which areas are encouraged to apply?\"] Students must be doing research work that relates to computing topics in which Microsoft Research is concerned with (click on Research Areas at the top of the page for a full list).\r\n\r\n<strong>Four major directions for reference:<\/strong>\r\n<ul>\r\n \t<li>Human machine intelligence\r\n<ul>\r\n \t<li>Big data mining<\/li>\r\n \t<li>Knowledge mining<\/li>\r\n \t<li>Machine learning<\/li>\r\n \t<li>Natural language processing<\/li>\r\n \t<li>Social computing<\/li>\r\n \t<li>Theory<\/li>\r\n \t<li>Urban computing<\/li>\r\n \t<li>Web search and data mining<\/li>\r\n \t<li>Others<\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Perception, recognition, interaction\r\n<ul>\r\n \t<li>Graphics<\/li>\r\n \t<li>Multimedia<\/li>\r\n \t<li>Robotics<\/li>\r\n \t<li>Speech<\/li>\r\n \t<li>Vision<\/li>\r\n \t<li>Others<\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Hardware and software systems\r\n<ul>\r\n \t<li>Applied algorithms<\/li>\r\n \t<li>Cloud computing<\/li>\r\n \t<li>Devices<\/li>\r\n \t<li>Mobile sensing<\/li>\r\n \t<li>Networking<\/li>\r\n \t<li>Software analytics<\/li>\r\n \t<li>Systems<\/li>\r\n \t<li>Others<\/li>\r\n<\/ul>\r\n<\/li>\r\n \t<li>Computational science\r\n<ul>\r\n \t<li>Computational ecology &amp; environmental studies<\/li>\r\n \t<li>Computational biomedical science &amp; public health<\/li>\r\n \t<li>Computational physics<\/li>\r\n \t<li>Computational materials science<\/li>\r\n \t<li>Other computational science fields<\/li>\r\n<\/ul>\r\n<\/li>\r\n<\/ul>\r\n<span style=\"line-height: 1.5\">[\/panel]<\/span>\r\n\r\n[panel header=\"Are first-year students eligible to apply?\"] First-year students are eligible to apply if they have already enrolled in a PhD program by the time of the nomination. [\/panel]\r\n\r\n[panel header=\"Are second-year students eligible to apply?\"] It depends on the expected graduation date of the normative PhD study you are enrolled in. If the student will graduate in 2023 or later, then they are eligible. [\/panel]\r\n\r\n[panel header=\"When will the review results be announced?\"] Selected applicants will receive notification of their acceptance status by early October 2021. Due to the volume of submissions, Microsoft Research Asia cannot provide individual feedback on applications that do not receive fellowship awards. [\/panel]\r\n\r\n[panel header=\"If selected, when can I come to Microsoft Research Asia for my internship?\"] A three-month internship at Microsoft Research Asia is guaranteed, and you may request to come for an internship at any time before you complete your PhD. Besides working with experienced mentors, an internship at Microsoft Research Asia allows fellows to present to and discuss with diverse researchers, which will help them expand their horizons and provide experience in large international corporations. To date, more than 200 fellows have interned. For more information about internship opportunities at Microsoft Research Asia, please visit the <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/careers\/\">Careers page<\/a>. [\/panel]\r\n\r\n[panel header=\"What are the intellectual property implications of a Microsoft Research Asia Fellowship?\"] The funds are given as an unrestricted gift. Fellowship recipients are not subject to intellectual property restrictions unless they complete an internship at Microsoft Research Asia. If that is the case, they are subject to the same intellectual property restrictions as any other Microsoft Research Asia intern. [\/panel]\r\n\r\n[panel header=\"Are students from other countries or regions eligible?\"] Microsoft Research Asia Fellowship only accepts applicants with universities in mainland China, Hong Kong, Japan, Korea, Singapore, and Taiwan.\r\n[\/panel]\r\n\r\n[\/accordion]\r\n\r\nPlease send email directly to <a href=\"mailto:fellowRA@microsoft.com\">fellowRA@microsoft.com<\/a> for other questions"},{"id":3,"name":"Fellows","content":"<h2>2021 Fellows<\/h2>\r\nOver one hundred fifty distinguished PhD candidates from 50 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2021 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 11 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.\r\n\r\nMicrosoft Research Asia recognizes the following fellows, who represent the best and the brightest PhD candidates in the Asia-Pacific region.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786601 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-dianyu-chen.jpg\" alt=\"Dianyu Chen\" width=\"170\" height=\"210\" \/>\r\n<h3><strong>Dianyu Chen<\/strong><\/h3>\r\nWestlake University\r\n\r\nSupervisors: <a href=\"https:\/\/sls.westlake.edu.cn\/en\/Faculty\/201912\/t20191202_2407.shtml\">Heping Xu<\/a>\r\n\r\n<strong>Research interests:<\/strong>\u00a0Computational Biology, Humoral Immunity\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0The humoral immune response is a type of adaptive immune response that enables the human body to defend itself in a targeted way, and is the basis for the vaccine to work. Despite considerable progress, there remain many knowledge gaps in the understanding of the cellular and molecular mechanisms underlying humoral immunity, which hinders the production of effective vaccines against many deadly viruses. The overall goal of my research is to leverage computational biology approaches to deepen our understanding about humoral immunity, thus providing new prescriptions for the development of urgently needed vaccines. Moreover, such knowledge can also contribute to the development of therapies for autoimmune disease.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786607 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-kaizhang-kang.jpg\" alt=\"Kaizhang Kang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/cocoakang.cn\/\"><strong>Kaizhang Kang<\/strong><\/a><\/h3>\r\nZhejiang University\r\n\r\nSupervisors: <a href=\"http:\/\/hongzhiwu.com\">Hongzhi Wu<\/a>\r\n\r\n<strong>Research interests:<\/strong>\u00a0Appearance Acquisition, 3D scanning and Machine Learning\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0My long-term research goal is to answer this question: How to digitize arbitrary real objects with challenging appearance properties and complex geometry in an efficient and high-quality manner? In past years, my research reformulated the 3D modeling problem and proposed a new methodology to jointly optimize both acquisition &amp; reconstruction procedures. By this, we proposed a differentiable acquisition framework as well as a series of high-performance setups. Although we can efficiently digitize most daily objects with high quality now, several properties remain hard to model, such as transparency. I am going to tackle these challenging problems in the future. In addition, enhancing medical imaging techniques by our novel acquisition framework is also an interesting topic to me.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786610 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-libo-qin.jpg\" alt=\"Libo Qin\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/ir.hit.edu.cn\/~lbqin\/\"><strong>Libo Qin<\/strong><\/a><\/h3>\r\nHarbin Institute of Technology\r\n\r\nSupervisor: <a href=\"http:\/\/ir.hit.edu.cn\/~car\/\">Wanxiang Che <\/a>\r\n\r\n<strong>Research interests<\/strong>: Natural Language Processing and Task-oriented Dialogue System\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0My research mainly focuses on natural language processing and task-oriented dialogue system (ToDs), which help users to achieve specific goals. With the burst of deep neural networks and the evolution of pre-trained language models, the research of ToDs has obtained significant breakthroughs. However, the existing models heavily rely on large training data, which is only available for a certain number of task domains or languages, lacking scalability, practicality, and robustness. Inspired by the great success of transfer learning, my long-term research goal is to explore transfer learning for ToDs, aiming to build a scalable, practical and robust task-oriented dialogue system with limited data.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786613 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qing-wang.jpg\" alt=\"Qing Wang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/storage.cs.tsinghua.edu.cn\/~wq\/\"><strong>Qing Wang<\/strong><\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: Jiwu Shu, Youyou Lu\r\n\r\n<strong>Research interests<\/strong>: Memory\/Storage Systems\r\n\r\n<strong>Long-term research goal<\/strong>: Shared memory systems underpin modern applications using a unified memory abstraction. My research rethinks shared memory systems, which is motivated by three hardware trends in datacenters, including new memory (e.g., persistent memory), new network (e.g., RDMA and CXL), and new compute (e.g., programmable switch ASIC). I have built several efficient shared memory systems at different system scales. My long-term research goal is abstracting the entire datacenter as a giant shared-memory machine, which can store, retrieve, and process data with high throughput, low latency, and good programmability.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786592 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-aoyu-wu.jpg\" alt=\"Aoyu Wu\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/wowjyu.github.io\/\"><strong>Aoyu Wu<\/strong><\/a><\/h3>\r\nHong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/huamin.org\/\">Huamin Qu<\/a>\r\n\r\n<strong>Research interests<\/strong>: Data Science and Visualization, Human-Computer Interaction\r\n\r\n<strong>Long-term research goal<\/strong>: Data visualizations often serve as the main entry point for the public to access data. My goal is to make visualizations a \u201cfirst-class citizen\u201d on the web and more accessible to both humans and computers. Specifically, I hope to build tools that automatically recommend models and visualizations for data analysis to advance the democratization of data science. Furthermore, unlike humans, machines have no direct access to the information inside visualizations. Massive numerical information and knowledge remain locked inside visualizations. I plan to create methods for interpreting and understanding the implicit meanings of visualizations at the Internet scale. By integrating those two perspectives, I aim to contribute to a new online knowledge ecology \u2013 both analyzing web visualizations to distill knowledge and assisting the public in producing new visualizations to communicate information. This ecology will increase the breadth of the semantic web and make \u201cbig data\u201d more accessible and efficient.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786616 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-qitian-wu.png\" alt=\"Qitian Wu\" width=\"170\" height=\"210\" \/>\r\n<h3><strong>Qitian Wu<\/strong><\/h3>\r\nHong Kong University of Science and Technology\r\n\r\nSupervisor: Junchi Yan\r\n\r\n<strong>Research interests:<\/strong>\u00a0Machine Learning, Data Mining\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0Modern machine learning systems need to interact with new unseen instances from a dynamic open world when adapted from training data to test observations or transferring from simulated environments to real situations. This requires the model\u2019s capability for in-the-wild generalization and extrapolation to new distributions. My research will focus on developing the next generation of representation learning approach that inherits the adaptability of human beings and can achieve inductive reasoning from familiar concepts to new ones. Here are two challenging problems to be resolved: how to define the relations among concepts and how to infer the latent structures among them. Such a new paradigm could benefit various tasks that need to obtain instance-specific embedding in broad AI areas and also shorten the gap between machines and human intelligence.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786604 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-hiromu-yakura.jpg\" alt=\"Hiromu Yakura\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/yumetaro.info\/\"><strong>Hiromu Yakura<\/strong><\/a><\/h3>\r\nUniversity of Tsukuba\r\n\r\nSupervisors: <a href=\"https:\/\/staff.aist.go.jp\/m.goto\/\">Masataka Goto<\/a>\r\n\r\n<strong>Research interests:<\/strong> Human-Computer Interaction, Machine Learning\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0As a researcher in HCI, my research focus is on expanding the application area of computers, especially, machine learning techniques. Given that we cannot deny the possibility of the mistakes of machine learning models, we need to carefully design the interaction between the models and humans to realize practical applications. In this context, I believe that understanding the characteristics of both machine learning models and humans is crucial. This is why I have associated my research with other interdisciplinary fields such as behavioral economics and cultural studies. I hope that I can reach a new perspective of understanding humans through exploring practical applications of machine learning in this human-centric manner.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786598 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-da-yu.jpg\" alt=\"Da Yu\" width=\"170\" height=\"210\" \/>\r\n<h3><strong>Da Yu<\/strong><\/h3>\r\nSun Yat-sen University\r\n\r\nSupervisors: Jian Yin, Tie-Yan Liu\r\n\r\n<strong>Research interests:<\/strong>\u00a0Privacy-preserving Learning, Trustworthy Deep Learning\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0My research aims to make deep learning safer to use. Despite its remarkable success, deep learning is prone to various of attacks, e.g., privacy attacks that steal user data and adversarial attacks that manipulate model outputs. Such attacks arouse concerns about deep learning and hinder its applications in real world. My future research goal is to build reliable algorithms to earn the trust of the participators from different aspects of deep learning.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786622 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wangyou-zhang.jpg\" alt=\"Wangyou Zhang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/x-lance.sjtu.edu.cn\/members\/wangyou-zhang\"><strong>Wangyou Zhang<\/strong><\/a><\/h3>\r\nShanghai Jiao Tong University\r\n\r\nSupervisor: Yanmin Qian\r\n\r\n<strong>Research interests:<\/strong>\u00a0Speech Signal Processing, Robust Speech Recognition\r\n\r\n<strong>Long-term research goal:<\/strong>\u00a0Speech is one of the most important ways in our daily communication. And \u201cone of our most important faculties is our ability to listen to, and follow, one speaker in the presence of others\u201d. However, the mechanism behind this ability and how machines can mimic human\u2019s ability in the complex real-world scenarios are still not well studied. This is known as the cocktail party problem, one of the most challenging problems in speech processing over the past 60 years. The advances in deep learning have greatly pushed forward the study in speech processing in the past decade. However, the cocktail party problem is still very difficult and remains unsolved. My long-term research goal is to address the multi-speaker speech modeling and processing in realistic scenarios and ultimately tackle the cocktail party problem.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786625 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-wanjun-zhong.jpg\" alt=\"a young man wearing a suit and tie\" width=\"170\" height=\"210\" \/>\r\n<h3><strong>Wanjun Zhong<\/strong><\/h3>\r\nSun Yat-Sen University\r\n\r\nSupervisors: Ming Zhou, Jian Yin, Jiahai Wang\r\n\r\n<strong>Research interests:<\/strong>\u00a0Machine Reasoning, Natural Language Processing\r\n\r\n<strong>Long-term research goal<\/strong>: Current neural network models learn from large-scale data, and its working mechanism is a black box. Therefore, they highly depend on large-scale annotated data, and are unreliable because the prediction is hard to be explained. Besides, it has limited use of human knowledge. To alleviate these problems, my research goal focuses on building reliable, data-efficient, controllable AI systems utilizing world knowledge. Finally, I hope the machines can better assist human to solve difficult reasoning tasks with world knowledge, like debates, reliable medical diagnosis, or solving mathematical problems.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-786619 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/10\/2021-fellows-tiankuang-zhou.jpg\" alt=\"Tiankuang Zhou\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/rickyim.github.io\/\"><strong>Tian<\/strong><strong>k<\/strong><strong>uang Zhou<\/strong><\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisors: <a href=\"http:\/\/www.luvision.net\/\">Lu Fang<\/a>, <a href=\"http:\/\/media.au.tsinghua.edu.cn\/\">Qionghai Dai<\/a>\r\n\r\n<strong>Research interests:<\/strong>\u00a0Optical Intelligent Computing, Machine Learning\r\n\r\n<strong>Long-term research goal<\/strong>: As the size of electronic transistor is approaching to its physical limit, electronic computer faces challenges in continuously improving the computing speed and energy efficiency. Modern computer science problems, however, requires drastically-increasing amount of compute. Endowed with high-speed and low-loss propagation, light is regarded as a promising candidate to remedy the conflict. My research focuses on employing photonics and optics to build novel computing architectures to solve computer science problems. My long-term goal is three-fold. Firstly, I want to build photonic computing machines that have the advantage to accelerate the inference process of artificial intelligence (AI). Secondly, since training of AI is time- and energy- consuming, I plan to construct efficient photonic systems for the training of AI. Thirdly, the computing process based on photons is fundamentally distinct from electrons. Because of this, I am striving for a photonic computing paradigm which can help us tackle the long-standing hard problems.\r\n\r\n<hr \/>\r\n\r\n&nbsp;\r\n\r\n[accordion]\r\n[panel header=\"2020 fellows\"]\r\nOver one hundred distinguished PhD candidates from 36 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2020 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 12 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.\r\n\r\nMicrosoft Research Asia recognizes the following fellows, who represent the best and the brightest PhD candidates in the Asia-Pacific region.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708853 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghyun-Woo_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Sanghyun Woo\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/sites.google.com\/view\/sanghyunwoo\/\" target=\"_blank\" rel=\"noopener\">Sanghyun Woo<\/a><\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: In So Kweon\r\n\r\n<b>Research interests:<\/b> Computer Vision, Machine Learning\r\n\r\n<b>Long-term research goal:<\/b> Artificial Intelligence (AI) is on its way to ubiquity. There has been a flurry of advances in AI technologies, yet the current AI systems are still too narrow and passive than human-level intelligence. My research aims to build an AI agent that can think and learn like humans. To do so, I focus on challenging issues in the current learning system both in the 'model' and 'data' perspective. Specifically, there are two principal directions I have explored. First, apart from the current model design paradigm, I attempted to incorporate the attention mechanism, a central component of the human vision system, into the model. Second, to reduce expensive and error-prone human supervision, I tried to exploit self-supervision or simulation and scale the training data up to a billion-scale. Based on my accumulated research experiences in both aspects, the future directions aim to unify two distinct perspectives and present the general learning framework that can incubate various future AI technologies.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708874 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Hao-Shu-Fang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Hao-Shu Fang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/fang-haoshu.github.io\/\" target=\"_blank\" rel=\"noopener\">Hao-Shu Fang<\/a><\/h3>\r\nShanghai Jiao Tong University\r\n\r\nSupervisor: <a href=\"http:\/\/mvig.sjtu.edu.cn\/\" target=\"_blank\" rel=\"noopener\">Cewu Lu<\/a>\r\n\r\n<b>Research interests:<\/b> Human Understanding, Robotics Grasping\/Manipulation\r\n\r\n<b>Long-term research goal:<\/b> Imagine a world where robots can learn to cook a perfect steak by observing your demonstration. My life dream is to build robots that are smart enough to learn many of the things humans do: assembly, cooking or cleaning the table. And I envision robots to quickly learn manipulation skills from human demonstration. This requires a breakthrough in several fields of AI, including human understanding and robotics manipulation. My research goal focuses on answering the following questions: <em>How can machines acquire visual, geometric and physical attributes of humans and objects? How to obtain an appropriate representation of human\u2019s demonstration? How to jointly model visual and force perception for robot manipulation? How to define an objective function for a robot?<\/em> These interesting and challenging problems motivate me to go further in my research.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708868 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Daya-Guo_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Daya Guo\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/guoday.github.io\/\" target=\"_blank\" rel=\"noopener\">Daya Guo<\/a><\/h3>\r\nSun Yat-sen University\r\n\r\nSupervisors: <a href=\"http:\/\/sdcs.sysu.edu.cn\/content\/2505\" target=\"_blank\" rel=\"noopener\">Jian Yin<\/a>,\u202f<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/mingzhou\/\">Ming Zhou<\/a> (Microsoft)\r\n\r\n<b>Research interests:<\/b> Natural Language Processing and Code Intelligence\r\n\r\n<b>Long-term research goal:<\/b> My research mainly focuses on natural language processing and code intelligence to enable computers to intelligently process, understand and generate both natural language and programming language. With the growing population of developers, leveraging AI technology to increase the productivity of developers becomes more and more important in both communities of software engineering and artificial intelligence. Inspired by the great success of large pre-trained models like BERT and GPT in natural language processing, my long-term research goal is to develop powerful pre-trained models to support various code-related tasks, e.g. code search and code completion, to improve the productivity of the development process.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708880 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Ling-Pan_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Ling Pan\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/ling-pan.github.io\/\" target=\"_blank\" rel=\"noopener\">Ling Pan<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/people.iiis.tsinghua.edu.cn\/~huang\/\" target=\"_blank\" rel=\"noopener\">Longbo Huang<\/a>\r\n\r\n<b>Research interests:<\/b> Reinforcement Learning, Computational Sustainability\r\n\r\n<b>Long-term research goal:<\/b> Recent years have witnessed great success of reinforcement learning (RL) with deep feature representations in many challenging tasks, including computer games, robotics, etc. However, it still faces several major challenges, especially when combined with deep neural networks. My research focuses on developing robust, efficient, and practical RL algorithms, and targets three core questions. Firstly, how can we ensure a convergent and robust learning behavior of an RL agent. Secondly, how can we tackle the high sample complexity problem of RL algorithms. Thirdly, how to successfully apply RL algorithms in important practical applications, e.g., computational sustainability problems.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708850 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Sanghoon-Kang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Sanghoon Kang\" width=\"170\" height=\"210\" \/>\r\n<h3>Sanghoon Kang<\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: Hoi-Jun Yoo\r\n\r\n<b>Research interests:<\/b> On-Device AI, SW\/HW Co-Design\r\n\r\n<b>Long-term research goal:<\/b> Despite the mind-blowing abilities of AI algorithms, it is still difficult to deploy AI applications to people\u2019s everyday devices. The inevitable physical limitation of data gathering prevents AI models from performing at its best in different environments. On-device training, which can solve the adaptability issue, is strongly challenged by the limited hardware performance and battery constraints of edge devices. My research goal is to develop software &amp; hardware approaches for energy-efficient AI computing, to enable edge devices with on-device training capabilities.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708877 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Juheon-Yi_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Juheon Yi\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/juheonyi.github.io\/\" target=\"_blank\" rel=\"noopener\">Juheon Yi<\/a><\/h3>\r\nSeoul National University\r\n\r\nSupervisor: <a href=\"http:\/\/youngkilee.blogspot.com\/\" target=\"_blank\" rel=\"noopener\">Youngki Lee<\/a>\r\n\r\n<b>Research interests:<\/b> Mixed Reality (MR), Mobile\/Embedded Deep Learning Systems\r\n\r\n<b>Long-term research goal:<\/b> The goal of my research is to design futuristic MR apps, comprehensively analyze their workloads, and build novel enabling mobile systems to support them. Despite their vast potential, truly immersive MR apps are yet to be realized. The core challenge lies in the unique workload of seamlessly combining the virtual contents over the physical world on resource-constrained mobile\/wearable devices. So far, I have been working on projects to develop core mobile\/embedded deep learning systems for fully-immersive MR apps. To realize my vision, I plan to cover more diverse aspects of MR by conducting interdisciplinary research at the intersection of mobile computing, deep learning, multimedia, and HCI.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708883 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Mo-Zou_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Mo Zou\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/ipads.se.sjtu.edu.cn\/pub\/members\/mo_zou\" target=\"_blank\" rel=\"noopener\">Mo Zou<\/a><\/h3>\r\nShanghai Jiao Tong University\r\n\r\nSupervisor:\u202fHaibo Chen\r\n\r\n<b>Research interests:<\/b> Formal Verification, Operating System\r\n\r\n<b>Long-term research goal:<\/b> System software such as operating systems is the backbone of computer systems. A single bug in operating systems could endanger the whole system. Formal verification can provide the highest level of assurance to system software. But applying formal verification to system software, especially concurrent ones, faces many challenges, from both theory and practice. My research tries to solve these challenges. My research goal is to make critical software systems truly reliable and secure through formal verification.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708862 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Yangbangyan-Jiang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Yangbangyan Jiang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/jiangyangby.github.io\/\" target=\"_blank\" rel=\"noopener\">Yangbangyan Jiang<\/a><\/h3>\r\nUniversity of Chinese Academy of Sciences\r\n\r\nSupervisor: <a href=\"https:\/\/qmhuang-ucas.github.io\/\" target=\"_blank\" rel=\"noopener\">Qingming Huang<\/a>\r\n\r\n<b>Research interests:<\/b> Deep Learning, Clustering\r\n\r\n<b>Long-term research goal:<\/b> Clustering provides us a way to release human efforts in annotation. Nowadays, we have witnessed the tremendous progress of clustering. However, the vast majority of existing methods only deal with the perfect data, e.g., the data are clean and balanced from distributional perspective, or have complete meta-information from structural perspective. To enhance their practicality, my long-term research goal is to improve the clustering methods in more challenging scenarios where the observed data are imperfect. Specifically, I plan to explore the correlation among the data distribution and structure, and design more accurate and more efficient clustering approaches on imperfect data.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708847 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Naoki-Kimura_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Naoki Kimura\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/nao-ki-mura.com\/\" target=\"_blank\" rel=\"noopener\">Naoki Kimura<\/a><\/h3>\r\nThe University of Tokyo\r\n\r\nSupervisor: <a href=\"https:\/\/lab.rekimoto.org\/\" target=\"_blank\" rel=\"noopener\">Junichi Rekimoto<\/a>\r\n\r\n<b>Research interests:<\/b> Human-Computer Interaction, Wearable Computing, Pattern Recognition\r\n\r\n<b>Long-term research goal:<\/b> I am developing an interface that allows every human being to use a computer without any special ability or training, like internal organs controlled by autonomic nerves. We still have many barriers to use computers. To use current computer systems, we need not only training, but also, at least vision and hands, or hearing and voice. The training keeps many people away from computers. Not all people have vision or hearing enough. Computers remain a tool for privileged people. For this goal, I have been working on Silent Speech Interfaces.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708856 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Tianyu-Pang_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Tianyu Pang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~tianyu\/\" target=\"_blank\" rel=\"noopener\">Tianyu Pang<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor:\u202f<a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~jun\/index.shtml\" target=\"_blank\" rel=\"noopener\">Jun Zhu<\/a>\r\n\r\n<b>Research interests:<\/b> Robust Machine Learning, Deep Generative Model\r\n\r\n<b>Long-term research goal:<\/b> My research interests span the areas of machine learning and deep learning, especially trust-worthy machine learning which deals with the problem of adversarial examples and more general robustness. The adversarial examples are maliciously crafted with human imperceptible perturbations, but can easily fool high-performance deep neural networks to return wrong decisions. The goal of my research is to design reliable, general strategies to enhance the model robustness, and explain the phenomena of adversarial vulnerability by the concepts from generative modeling or casual inference.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708859 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Xu-Han_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Xu Han\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/thucsthanxu13.github.io\/\" target=\"_blank\" rel=\"noopener\">Xu Han<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisors: Zhiyuan Liu, Maosong Sun\r\n\r\n<b>Research interests:<\/b> Natural Language Processing, Information Extraction, Knowledge Graph\r\n\r\n<b>Long-term research goal:<\/b> My research focuses on the intersection of natural language processing, information extraction, and knowledge graph. Up to now, various large-scale knowledge graphs have been constructed to organize world knowledge in a structured form, which can benefit many knowledge-driven applications. Although existing knowledge graphs may contain hundreds of millions of facts, they are still far away from completion and require further efforts to acquire new knowledge. The goal of my research is to extract knowledge from the text to form a complete knowledge graph and meanwhile utilize the knowledge graph to perform better language understanding.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-708865 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2020\/11\/Zili-Meng_170x210.jpg\" alt=\"Microsoft Research Asia 2020 Fellow: Zili Meng\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/zilimeng.com\" target=\"_blank\" rel=\"noopener\">Zili Meng<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/routing.netlab.edu.cn\/tiki-index.php?page=Mingwei+Xu\" target=\"_blank\" rel=\"noopener\">Mingwei Xu<\/a>\r\n\r\n<b>Research interests:<\/b> Networking Systems, Video Streaming, Deep Learning\r\n\r\n<b>Long-term research goal:<\/b> I conduct research in the area of computer networks, focusing on the intersection area between deep learning and networking system. When improving the performance of networking systems, existing deep learning techniques also introduce a series of drawbacks (e.g., heavyweight, noninterpretable, etc.). My long-term research goal is to make deep learning-based networking systems deployable in practice and enjoy the performance benefits without the drawbacks. Besides, I\u2019m also interested in the optimization of transport in video streaming with advanced techniques.\r\n\r\n[\/panel]\r\n[panel header=\"2019 fellows\"]\r\nOver one hundred distinguished PhD candidates from 31 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2019 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 12 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.\r\n\r\n[caption id=\"attachment_620817\" align=\"aligncenter\" width=\"1024\"]<img class=\"wp-image-620817 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/MSRA-fellows-2019_multi_11_2019_1400x788-1024x576.jpg\" alt=\"2019 Microsoft Research Asia Fellows\" width=\"1024\" height=\"576\" \/> Microsoft Research Asia 2019 fellows with Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&amp;D Group, Microsoft Research Asia (first from right in back row), and Tie-Yan Liu, Assistant Managing Director of Microsoft Research Asia (first from left in back row).[\/caption]\r\n<img class=\"wp-image-620865 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Shijie-Cao_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Shijie Cao\" width=\"170\" height=\"210\" \/>\r\n\r\n<h3>Shijie Cao<\/h3>\r\nHarbin Institute of Technology\r\n\r\nSupervisors: Lintao Zhang, Dechen Zhang\r\n\r\n<b>Research interests:<\/b> Deep Learning, High Performance Computing, Reconfigurable Computing\r\n\r\n<b>Long-term research goal:<\/b> My research interest lies at the intersection of deep learning and high-performance computing. Deep learning has revolutionized vision, language and speech fields with tremendous large and multi-layer deep neural networks. Inference on these bulky models often requires a huge amount of computational power. However, the slowing down of Moore\u2019s law and Dennard scaling has led to the end of rapid performance improvement in general-purpose processors. Running DNN models in a low-cost, low-latency and energy-ef\ufb01cient way is highly desirable and has attracted much attention in the research community. My long-term research goal is co-designing deep learning models\/algorithms and domain specific hardware architectures for efficient deep learning computing.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620874 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Zhenpeng-Chen_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Zhenpeng Chen\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/chenzhenpeng18.github.io\/\" target=\"_blank\" rel=\"noopener\">Zhenpeng Chen<\/a><\/h3>\r\nPeking University\r\n\r\nSupervisors: <a href=\"http:\/\/sei.pku.edu.cn\/~meih\/index_en.html\" target=\"_blank\" rel=\"noopener\">Hong Mei<\/a>, <a href=\"http:\/\/www.liuxuanzhe.com\/\" target=\"_blank\" rel=\"noopener\">Xuanzhe Liu<\/a>\r\n\r\n<b>Research interests:<\/b> Software Analytics, Web Mining\r\n\r\n<b>Long-term research goal:<\/b> To help software practitioners understand their target audience and thus obtain actionable insights for improving user experience, powerful techniques, such as user profiling and sentiment analysis, have been proposed. However, many of them are performed based on user-generated texts, mainly English texts. Although some efforts have also been made with other languages, related techniques for non-English languages are far behind. This creates a considerable inequality in the quality of the related services received by non-English users, especially considering that about 75% of Internet users are non-English speakers. To alleviate such an inequality, the long-term goal of my research is to develop novel approaches to understanding users through ubiquitous data such as emojis. Since these data are language-independent, techniques powered by them can benefit users in different languages.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620871 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Yinpeng-Dong_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Yinpeng Dong\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~yinpeng\/\" target=\"_blank\" rel=\"noopener\">Yinpeng Dong<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~jun\/index.shtml\" target=\"_blank\" rel=\"noopener\">Jun Zhu<\/a>\r\n\r\n<b>Research interests:<\/b> Robust Deep Learning, Adversarial Machine Learning\r\n\r\n<b>Long-term research goal:<\/b> My research focuses on the robustness issue of machine learning and deep learning models in the adversarial setting. These models are shown to be significantly vulnerable to adversarial examples, which are maliciously crafted by adding small perturbations to the normal ones, but make the target model produce erroneous predictions. The goal of my research is to thoroughly understand the robustness of these models through adversarial attacks; develop more robust machine learning and deep learning models, and build a comprehensive benchmark to evaluate adversarial robustness in various settings.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620859 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Insu-Han_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Insu Han\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/insuhan.github.io\/\" target=\"_blank\" rel=\"noopener\">Insu Han<\/a><\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/alinlab.kaist.ac.kr\/shin.html\" target=\"_blank\" rel=\"noopener\">Jinwoo Shin<\/a>\r\n\r\n<b>Research interests:<\/b> Large-scale Machine Learning, Approximate Algorithm\r\n\r\n<b>Long-term research goal:<\/b> In many modern machine learning problems, it is desirable to develop scalable algorithms suitable for large-scale settings. My research goal is to design fast and reliable approximate algorithms for a wide range of machine learning problems. In particular, I am interested in problems related to matrix spectral functions, for instance, optimizing recommendation systems, inference of diverse subset selection and so on. For future research, I will focus on developing fast but theoretically guaranteed approximate algorithms in fundamental machine learning problems so that they are indeed applicable in practical settings.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620853 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Hirofumi-Inaguma_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Hirofumi Inaguma\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/hirofumi0810.github.io\/\" target=\"_blank\" rel=\"noopener\">Hirofumi Inaguma<\/a><\/h3>\r\nKyoto University\r\n\r\nSupervisor: <a href=\"http:\/\/sap.ist.i.kyoto-u.ac.jp\/members\/kawahara\/home_e.html\" target=\"_blank\" rel=\"noopener\">Tatsuya Kawahara<\/a>\r\n\r\n<b>Research interests:<\/b> Automatic speech recognition, speech translation\r\n\r\n<b>Long-term research goal:<\/b> Speech translation (ST) is a key technique to break the language barrier in human communication. Automatic speech recognition (ASR) and machine translation (MT) are its essential components. However, error propagations from the ASR module prevent the following MT module from translating source text correctly and this results in conveying incorrect information. To address this, end-to-end ST (E2E-ST) has been investigated to directly translate source speech to target languages. The long-term goal of my research is to explore novel approaches to realize the universal multilingual E2E-ST system, which takes contextual information shared among multiple languages into account and generates target translations in an online streaming manner.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620850 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Dahun-KIm_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Dahun KIm\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/sunghoonim.github.io\/\" target=\"_blank\" rel=\"noopener\">Dahun KIm<\/a><\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/rcv.kaist.ac.kr\/\" target=\"_blank\" rel=\"noopener\">In So Kweon<\/a>\r\n\r\n<b>Research interests:<\/b> Computer Vision and Machine Learning\r\n\r\n<b>Long-term research goal:<\/b> We are in the midst of content explosion, surrounded by an endless supply of images and videos. My research aims at creating machines that can understand the visual content with minimal human supervision. This also challenges the biggest bottleneck in recent deep learning technologies: having humans annotate a large set of data is expensive, error-prone, and limits the scalability to real-world applications. I have explored algorithms that can learn with weak or self-supervision. The ongoing and future directions aim for the video domain, including video editing and understanding tasks, where the limitation of human annotation is even more severe.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620880 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/Chao_Liao_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Chao Liao\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/chaoliao.org\/\" target=\"_blank\" rel=\"noopener\">Chao Liao<\/a><\/h3>\r\nShanghai Jiao Tong University\r\n\r\nSupervisors: <a href=\"http:\/\/itcs.shufe.edu.cn\/pinyan\/\" target=\"_blank\" rel=\"noopener\">Pingyan Lu<\/a>, <a href=\"http:\/\/apex.sjtu.edu.cn\/members\/yyu\" target=\"_blank\" rel=\"noopener\">Yong Yu<\/a>\r\n\r\n<b>Research interests:<\/b> Theoretical Computer Science, Approximate Counting and Sampling\r\n\r\n<b>Long-term research goal:<\/b> Counting problems naturally arise in many subjects, including combinatorics, discrete probability, statistical physics and many others. From a computational perspective, we would like to obtain fast algorithms to count the number of certain objects (exactly or approximately), or prove that such algorithms do not exist. Over the past decades, many interesting results have been discovered. One highlight is the correspondence between the computational phase transition and the uniqueness of Gibbs measures on the hard-core model. Even though, our understanding of the whole picture is still far from complete. In the future, I plan to explore the connections between the existence of efficient counting algorithms and other subjects, hoping to provide new insights into the picture.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620862 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Muoi-Tran_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Muoi Tran\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/www.comp.nus.edu.sg\/~muoitran\/\" target=\"_blank\" rel=\"noopener\">Muoi Tran<\/a><\/h3>\r\nNational University of Singapore, School of Computing\r\n\r\nSupervisor: <a href=\"https:\/\/www.comp.nus.edu.sg\/~kangms\/\" target=\"_blank\" rel=\"noopener\">Min Suk Kang<\/a>\r\n\r\n<b>Research interests:<\/b> Network Security, Blockchain Security\r\n\r\n<b>Long-term research goal:<\/b> Whether it is a multibillion-dollar cryptocurrency (e.g., Bitcoin) or a toy blockchain implemented for fun, the agreements among distributed nodes heavily rely on the underlying peer-to-peer network for data transmission. Therefore, this network has been an attractive target of attacks in recent years\u2014large Internet Service Providers have been shown to be capable of disrupting it effectively and stealthily. Currently, we focus on both uncovering such a family of attacks and building practical detection systems against them. Our long-term research goal is to design secure peer-to-peer networks that fundamentally harden thousands of cryptocurrencies, and even against strong network adversary model such as Internet Service Providers.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620868 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Taining-Wang_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Taining Wang\" width=\"170\" height=\"210\" \/>\r\n<h3>Taining Wang<\/h3>\r\nNational University of Singapore\r\n\r\nSupervisor: <a href=\"https:\/\/www.comp.nus.edu.sg\/~chancy\/\" target=\"_blank\" rel=\"noopener\">Chee-Yong Chan<\/a>\r\n\r\n<b>Research interests:<\/b> Database query processing and optimization, data sampling, data analytics\r\n\r\n<b>Long-term research goal:<\/b> My long-term research goal is to develop algorithms to improve the query processing and data analytics functionalities in data management systems. Managing and analyzing data efficiently and effectively has always been a key concern in data management systems. However, there are still many unsolved problems in efficient and effective data processing. One example of such problems is on optimizing the execution order of the join operators in a join query. This is a critical problem in database query optimization, because a good join ordering can usually lead to orders of magnitude improvement on the query execution time. The main difficulty is to accurately estimate the size of the join query result with minimal computation and storage overhead. This is one of the open problems in efficient and effective data processing among others.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620847 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Chuhan-Wu_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Chuhan Wu\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/wuch15.github.io\/\" target=\"_blank\" rel=\"noopener\">Chuhan Wu<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"https:\/\/www.tsinghua.edu.cn\/publish\/eeen\/3784\/2010\/20101219113838975735106\/20101219113838975735106_.html\" target=\"_blank\" rel=\"noopener\">Yongfeng Huang<\/a>\r\n\r\n<b>Research interests:<\/b> Recommender Systems, User Modeling and Natural Language Processing\r\n\r\n<b>Long-term research goal:<\/b> User modeling is important for understanding users and providing personalized services. However, existing user modeling methods usually rely on centralized storage of user behavior data and centralized model learning, which may bring a series of challenges on privacy, such as the privacy concerns of users, the difficulties to need data protect regulations and the risk of user data leakage. Thus, privacy-preserving user modeling is necessary. In the future, in addition to developing more accurate and efficient user modeling methods, I plan to explore how to model users and provide personalized services in a privacy-preserving manner.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620856 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Hongming-Zhang_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Hongming Zhang\" width=\"170\" height=\"210\" \/>\r\n<h3>Hongming Zhang<\/h3>\r\nThe Hong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"https:\/\/www.cse.ust.hk\/~yqsong\/\" target=\"_blank\" rel=\"noopener\">Yangqiu Song<\/a>\r\n\r\n<b>Research interests:<\/b> Commonsense Reasoning\r\n\r\n<b>Long-term research goal:<\/b> Commonsense reasoning, as one of the most challenging tasks in the natural language processing community, is crucial for machines to truly understand human language. In the future, I plan to focus on solving commonsense reasoning problems with structured knowledge. To be specific, I am trying to explore a structured way of representing commonsense knowledge, an efficient way of extracting commonsense knowledge, and a principle way of applying commonsense knowledge into downstream tasks.\r\n\r\n<hr \/>\r\n\r\n<img class=\"wp-image-620877 size-full alignleft\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2019\/11\/Zizhao-Zhang_170x210.jpg\" alt=\"Microsoft Research Asia - 2019 Fellow: Zizhao Zhang\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/zzz14.github.io\/\" target=\"_blank\" rel=\"noopener\">Zizhao Zhang<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/gaoyue.org\/cn_tsinghua\/people\/gaoyue_index.html\" target=\"_blank\" rel=\"noopener\">Yue Gao<\/a>\r\n\r\n<b>Research interests:<\/b> Brain Science, Graph Signal Processing, Complex Network\r\n\r\n<b>Long-term research goal:<\/b> In recent years, many brain disorders have been found to be associated with topological structure abnormalities of large-scale functional networks, such as Alzheimer's (AD), Parkinson\u2019s (PD), autism spectrum disorder (ASD) and schizophrenia (SZ). However, human brain is a complex network and the brain functional activity is a real-time dynamic process. How to bridge the gap between the brain networks and the brain disorders is an important task. The long-term goal of my research is to establish the automated diagnosis and pathological mechanisms understanding of brain disorders using brain functional and structural network.\r\n\r\n[\/panel]\r\n[panel header=\"2018 fellows\"] Over one hundred distinguished PhD candidates from 40 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2018 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 11 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.\r\n\r\n[caption id=\"attachment_551808\" align=\"aligncenter\" width=\"1024\"]<img class=\"wp-image-551808 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/1-2018-Fellow-with-Peter-and-Hon-eyes-1024x513.jpg\" alt=\"\" width=\"1024\" height=\"513\" \/> Microsoft Research Asia 2018 fellows with Peter Lee, CVP, Microsoft Research (1st from the right); Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&amp;D Group, Microsoft Research Asia (2nd from the left), and Lidong Zhou, Assistant Managing Director of Microsoft Research Asia (1st from the left).[\/caption]\r\n\r\n&nbsp;\r\n\r\n[caption id=\"attachment_551811\" align=\"aligncenter\" width=\"1024\"]<img class=\"wp-image-551811 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/2-2018-Fellow-with-Raj-Reddy-1024x521.jpg\" alt=\"\" width=\"1024\" height=\"521\" \/> After Award Ceremony, Microsoft Research Asia 2018 fellows had roundtable meeting with Raj Reddy, 1994 Turing Award Recipient.[\/caption]\r\n\r\n&nbsp;\r\n\r\n[caption id=\"attachment_551814\" align=\"aligncenter\" width=\"1024\"]<img class=\"wp-image-551814 size-large\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2016\/04\/3-2018-Fellow-with-Yao-1024x683.jpg\" alt=\"\" width=\"1024\" height=\"683\" \/> Microsoft Research Asia 2018 fellows had roundtable meeting with Andrew Chi-Chih Yao, 2000 Turing Award Recipient.[\/caption]\r\n\r\n&nbsp;\r\n\r\n<img class=\"alignleft wp-image-551820 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/1-Weiming-Feng.png\" alt=\"Weiming Feng - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/tcs.nju.edu.cn\/fengwm\/\" target=\"_blank\" rel=\"noopener\">Weiming Feng<\/a><\/h3>\r\nNanjing University\r\n\r\nSupervisors: <a href=\"http:\/\/tcs.nju.edu.cn\/yinyt\/Yitong_Yin_homepage.html\" target=\"_blank\" rel=\"noopener\">Yitong Yin<\/a>\r\n\r\n<b>Research interests:<\/b> Randomized Algorithms, Theory of Distributed Computing\r\n\r\n<b>Long-term research goal:<\/b> Sampling is an extensively studied topic in computer science and statistical physics. However, many classic and widely-used sampling algorithms such as single-site dynamics are fully adaptive and highly sequential. Nowadays, the size of data set grows rapidly so that the sequential sampling algorithms become inefficient. As a result, the distributed sampling algorithms attract more and more attention, especially on the area of distributed machine learning. However, the theory of distributed sampling is lacking of systematic studies. The long-term goal of my research is to propose novel algorithms and provide a better understanding of distributed sampling.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551823 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/2-Sunghoon-Im.png\" alt=\"Sunghoon Im - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/sunghoonim.github.io\/\">Sunghoon Im<\/a><\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/rcv.kaist.ac.kr\/index.php?mid=rcv_faculty\">In So Kweon<\/a>\r\n\r\n<strong>Research interests: <\/strong>Computer Vision and Machine Learning\r\n\r\n<strong>Long-term research goal: <\/strong>3D scene understanding is essential for AR\/VR application and robotics application; however, capturing 3D data requires costly and bulky additional devices that reduce commercial use. My research goal is to provide a cost and size efficient solution for visual tracking and reconstruction in the hand-held device industry through the combination of modern geometric knowledge and machine learning techniques. In the future, I plan on exploring additional novel approaches that can <u>be applied<\/u> to commercial products.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551826 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/3-Dongup-Kwon.png\" alt=\"Dongup Kwon - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/hpcs.snu.ac.kr\/~dongup\/\" target=\"_blank\" rel=\"noopener\">Dongup Kwon<\/a><\/h3>\r\nSeoul National University\r\n\r\nSupervisor: <a href=\"https:\/\/hpcs.snu.ac.kr\/~jangwoo\/\" target=\"_blank\" rel=\"noopener\">Jangwoo Kim<\/a>\r\n\r\n<strong>Research interests:<\/strong> Server Architecture, Accelerator Architecture, System Software\r\n\r\n<strong>Long-term research goal:<\/strong> My research interests are in developing and evaluating computer systems targeting emerging server applications such as data analytics, machine learning, and brain simulations. Since such workloads demand much higher computational capabilities at unprecedented scales, it is critical to design extremely fast and scalable server systems. My research goal is to develop device-centric and extreme-scale server systems for large-scale and distributed workloads. I mainly focus on HW\/SW co-optimization and FPGA\/ASIC-based accelerators. I firmly believe that my research topic will have a significant impact on modern domain-specific or cloud-based computing paradigms.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551829 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/4-Yikang-Li.png\" alt=\"Yikang Li - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/cvboy.com\/\" target=\"_blank\" rel=\"noopener\">Yikang LI<\/a><\/h3>\r\nThe Chinese University of Hong Kong\r\n\r\nSupervisor: Xiaogang WANG\r\n\r\n<strong>Research interests:<\/strong> Deep Learning, Computer Vision, Vision and Language\r\n\r\n<strong>Long-term research goal:<\/strong> With the development of the Computer Vision (CV) and Natural Language Processing (NLP), the intersection area, Vision and Language, has drawn increasingly more attentions these years. Integrating language with vision brings with it the possibility of expanding the horizons and tasks of the vision community. We have seen significant growth in image\/video-to-text tasks but many other potential applications of such integration \u2013 answering questions, dialog systems, and grounded language acquisition \u2013 remain largely unexplored. Going beyond such novel tasks, language can make a deeper contribution to vision: it provides a prism through which to understand the world. Therefore, my long-term goal is to make the computer Understanding and Express the visual world in a more human-friendly way by integrating the NLP research.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551850 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/5-Jiaxin-Shi.png\" alt=\"Jiaxin Shi - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3>Jiaxin Shi<\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/ml.cs.tsinghua.edu.cn\/~jun\/index.shtml\" target=\"_blank\" rel=\"noopener\">Jun Zhu<\/a>\r\n\r\n<strong>Research interest:<\/strong> Probabilistic Learning, Bayesian Inference, Bayesian Deep Learning\r\n\r\n<strong>Long-term Research Goal:<\/strong> The main theme of my research is to address fundamental problems from probabilistic machine learning, Bayesian methods, and challenges from emerging fields like Bayesian deep learning. The goal is to increase the flexibility of probabilistic models, while keeping their inference and learning easy, and sufficiently scalable to solve real-world problems. Examples include unsupervised (semi-supervised) learning, structured prediction, and uncertainty estimation in supervised learning. I created and lead the development of ZhuSuan, a probabilistic programming library for Bayesian deep learning.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551832 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/6-Lili-Wei.png\" alt=\"Lili Wei - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/home.cse.ust.hk\/~lweiae\/\" target=\"_blank\" rel=\"noopener\">Lili WEI<\/a><\/h3>\r\nThe Hong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/www.cse.ust.hk\/faculty\/scc\/\" target=\"_blank\" rel=\"noopener\">Shing-Chi Cheung<\/a>\r\n\r\n<strong>Research interests:<\/strong> Software engineering: program analysis and testing for mobile applications\r\n\r\n<strong>Long-term research goal:<\/strong> The growing popularity of mobile devices has emerged the need of quality assurance for mobile applications. I have been focused on automatically identifying bugs in mobile applications. One of my major research projects is to characterize and detect compatibility issues induced by Android fragmentation. Due to its evolving and open nature, Android ecosystem is heavily fragmented. Various compatibility issues thus arise in Android applications. This has been well-recognized as one of the biggest challenges in Android application development. I have conducted studies to characterize and automatically detect these fragmentation-induced compatibility issues. In long term, I aim to develop a tool chain that can help app developers to automatically detect, diagnose and fix these compatibilities issues.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551835 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/7-Xingda-Wei.png\" alt=\"Xingda Wei - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3>Xingda Wei<\/h3>\r\nShanghai Jiaotong University\r\n\r\nSupervisor: Rong Chen, Haibo Chen and Bingyu Zang\r\n\r\n<strong>Research interests:<\/strong> Distributed systems and storage systems\r\n\r\n<strong>Long-term research goal:<\/strong> Distributed transactional system is a key component for many data-center applications, which serves the storage backend for applications such as e-commerce applications, social networks and websites. However, efficient processing transactions are hard, especially in a distributed setting. My research is to improve the performance and reliability of distributed systems including distributed transactions. I mainly use two approaches to achieve this. The first is through an algorithmic way, which use algorithms to achieve better application properties (e.g. use fewer network communications) given common cases. The second is through hardware and system co-design, which integrate advanced hardware (like RDMA and HTM) into system designs. To this end, our innovations in distributed transactions can provide orders of magnitude better performance than priori solutions.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551838 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/8-Lijun-wu.png\" alt=\"Lijun Wu - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3>Lijun Wu<\/h3>\r\nSun Yat-sen University\r\n\r\nSupervisor: Jianhuang Lai and Tie-Yan Liu\r\n\r\n<strong>Research interests:<\/strong> Machine Learning, Reinforcement Learning, Neural Machine Translation\r\n\r\n<strong>Long-term research goal:<\/strong> Neural machine translation (NMT) has drawn more and more attention in both academia and industry. With the development of deep learning and strong machine learning methods, NMT has achieved near human-level performance in some language pairs. However, current state-of-the-art NMT models require large amounts of parameters and still take several days to achieve a high performance. Besides, the fixed model structure and loss function make the NMT learning problem not easy to make a big breakout. My research goal is to design light and efficient machine learning models for machine translation, and eventually try to automatically \u201clearning to teach\u201d the optimal model for NMT or other NLP tasks.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551841 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/9-Fengli-Xu.png\" alt=\"Fengli Xu - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/fenglixu.com\/\" target=\"_blank\" rel=\"noopener\">Fengli Xu<\/a><\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/fi.ee.tsinghua.edu.cn\/~liyong\/\" target=\"_blank\" rel=\"noopener\">Yong Li<\/a>\r\n\r\n<strong>Research interests:<\/strong> Data-driven Human Behavior Modelling and Data Privacy\r\n\r\n<strong>Long-term research goal:<\/strong> The dramatic proliferation of smart devices has significantly contributed to the increasingly available fine-grained human behavioral big data. Such datasets have enormous potential in pushing forward the frontier of human behavior modelling, but they also pose challenges to current analytic frameworks and user privacy preservation techniques. My research contributes to the endeavor of harnessing the power of human behavioral big data in two ways: developing compatible analytic frameworks and proposing privacy models that allow datasets to safely circulate. My long term research goal is to expand my current research to facilitate advanced privacy-preserving human-centric computing.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551844 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/10-Pan-Zhou.png\" alt=\"Pan Zhou - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3>Pan Zhou<\/h3>\r\nNational University of Singapore\r\n\r\nSupervisor: Jiashi Feng and Shuicheng Yan\r\n\r\n<strong>Research interests:<\/strong> Structural Data Analysis, Optimization Algorithm, Deep Learning Theory\r\n\r\n<strong>Long-term research goal:<\/strong> Data modeling method and optimization algorithm are two key factors for achieving success in real applications. For instance, thanks to the strong data modeling ability of deep learning and the effectiveness of stochastic backpropagation algorithm, we have achieved remarkable success in many AI fields. So like my previous work, in the future I continue to focus my research interest on developing stronger data modeling methods and more efficient optimization algorithms, as well as establishing their theoretical performance guarantees. Especially, I will put more my efforts on designing effective network architecture and developing more efficient training algorithm in deep learning because of its favorable usage in applications. My future goal is to provide a new tool, including a data modeling technique and its algorithm, for better modeling the real data with high efficiency.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-551847 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/11\/11-Xizhou-Zhu.png\" alt=\"Xizhou Zhu - Fellowships at Microsoft Research Asia\" width=\"170\" height=\"210\" \/>\r\n<h3>Xizhou Zhu<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisor: <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/bainguo\/\">Baining Guo<\/a> and <a href=\"http:\/\/staff.ustc.edu.cn\/~xjchen99\/\" target=\"_blank\" rel=\"noopener\">Xuejin Chen<\/a>\r\n\r\n<strong>Research interests:<\/strong> Deep Learning, Video Recognition\r\n\r\n<strong>Long-term research goal:<\/strong> Recent years have witnessed significant success of deep convolutional neutral networks for image recognition. With their success, the recognition tasks have been extended from image domain to video domain. Fast and accurate video recognition is crucial for high-value scenarios, e.g., autonomous driving and video surveillance. I have proposed flow-based methods to exploit motion for video recognition tasks, which has achieved both faster speed and better accuracy. These methods share the same principles: motion estimation module built into the network architecture and end-to-end learning of all modules performed over multiple frames. I insist on principled multi-frame feature learning instead of heuristics, which is general for different tasks. For future research, I will focus on not only better speed-accuracy tradeoff but also challenges beyond speed and accuracy for video recognition, e.g. low latency and stability. [\/panel] [panel header=\"2017 fellows\"]\r\n<h2>2017 fellows announced<\/h2>\r\nMore than 100 distinguished PhD candidates from 38 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore have applied for the 2017 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 10 outstanding PhD candidates whose exceptional talent and innovation in computer science\u2013related research identifies them as emerging leaders in the Asia-Pacific region.\r\n\r\n[caption id=\"attachment_435546\" align=\"aligncenter\" width=\"808\"]<img class=\" wp-image-435546\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/2017_MSRA_Fellows_Peter_Hon.jpg\" alt=\"Microsoft Research Asia fellows 2017\" width=\"808\" height=\"442\" \/> Microsoft Research Asia 2017 fellows with Peter Lee, CVP, Microsoft Research (far right); Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&amp;D Group, Microsoft Research Asia (second from the left); and Tim Pan, Senior Outreach Director of Microsoft Research Asia (far left).[\/caption]\r\n\r\n[caption id=\"attachment_435543\" align=\"aligncenter\" width=\"808\"]<img class=\" wp-image-435543\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/2017-Fellow-with-John-Hopcroft.jpg\" alt=\"Microsoft Research Asia fellows 2017\" width=\"808\" height=\"444\" \/> After the award ceremony, Microsoft Research Asia 2017 fellows had roundtable meeting with John Hopcroft, 1986 Turing Award Recipient.[\/caption]\r\n\r\nMicrosoft Research Asia recognizes the following fellows, who represent the best and the brightest PhD candidates in the Asia-Pacific region.\r\n\r\n&nbsp;\r\n\r\n<img class=\"alignleft wp-image-435561 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Seung-Hwan-Baek.jpg\" alt=\"Seung-Hwan Baek Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/sites.google.com\/view\/shbaek\/\" target=\"_blank\" rel=\"noopener\">Seung-Hwan Baek<\/a><\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/vclab.kaist.ac.kr\/minhkim\/index.html\" target=\"_blank\" rel=\"noopener noreferrer\">Min H. Kim<\/a>\r\n\r\n<b>Research interests:<\/b> Computational imaging and computational photography\r\n\r\n<b>Long-term research goal:<\/b> Visual data is key to understanding the real world for both humans and robots. However, due to the limitations of current imaging systems, capturing the real world in full fidelity is still challenging. My research goal is to overcome these limitations by designing computational imaging systems based on computer graphics, computer vision, and optics. I take a multidisciplinary approach, not only designing a problem in a mathematical form, but also exploring diverse optical phenomena to reveal invisible properties of the real world: depth, material reflectance, and spectrum. I envision that my research will provide new tools for better understanding the real world, which is essential for many applications such as VR\/AR and autonomous driving.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435570 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Yue-Cao.jpg\" alt=\"Yue Cao Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Yue Cao<\/h3>\r\nTsinghua University\r\n\r\nSupervisors: Jianmin Wang and Mingsheng Long\r\n\r\n<b>Research interests:<\/b> Deep learning and computer vision\r\n\r\n<b>Long-term research goal:<\/b> Due to the tremendous increase in the size of data with high dimensions, it remains a great challenge to efficiently search for multimedia data. To guarantee both retrieval quality and computation efficiency, the approximate nearest neighbor (ANN) search has attracted increasing attention and is shown to be very useful for many practical problems, such as similar image search, object retrieval, cross-modal retrieval, and more. In the future, I\u2019ll continue to explore novel approaches to further improve the accuracy and efficiency of approximate nearest neighbor search.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435549 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Bojie-Li.jpg\" alt=\"Bojie Li Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Bojie Li<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisors: Enhong Chen and Lintao Zhang\r\n\r\n<b>Research interests:<\/b> Networked systems and reconfigurable hardware\r\n\r\n<b>Long-term research goal:<\/b> I\u2019m pushing forward research in accelerating datacenter infrastructure through reconfigurable hardware. There is an increasing performance mismatch between CPU and specialized accelerators (such as GPU, TPU), storage, and network components. Because Microsoft is deploying reconfigurable hardware in the public cloud, it gives me the opportunity to explore this potentially fruitful research area, now in its infancy. Our solution is to offload common patterns in computation, communication, and coordination from CPU to programmable NIC, as well as rethinking the architecture and programming model for the network switch, NIC, OS, and applications. To this end, we have designed systems to achieve significant speedup in network functions, in-memory key-value store, distributed transactions, and container networking. I hope to glue these pieces of work into a coherent picture of datacenter infrastructure, and ultimately arrive at elementary abstractions to build efficient and scalable systems in the future.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435555 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Chongxuan-Li.jpg\" alt=\"Chongxuan Li Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Chongxuan Li<\/h3>\r\nTsinghua University\r\n\r\nSupervisors: Bo Zhang and Jun Zhu\r\n\r\n<b>Research interests:<\/b> Deep learning and deep generative models\r\n\r\n<b>Long-term research goal:<\/b> My research interests are primarily on deep generative models (DGMs), which conjoin the flexibility of deep neural networks and the inference power of generative approaches. Via probabilistic modelling and inference, DGMs can handle the uncertainty of the input data and extract meaningful features without supervision, which are key to building human-like AI. My long-term research goal is to develop novel DGMs for challenging learning tasks, especially semi-supervised DGMs on partially labelled data, structured DGMs on sequential data, and DGMs with decision making in reinforcement learning settings.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435555 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Pengfei-Liu.jpg\" alt=\"Pengfei Liu Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Pengfei Liu<\/h3>\r\nFudan University\r\n\r\nSupervisors: Xuanjing Huang and Xipeng Qiu\r\n\r\n<b>Research interests:<\/b> Natural language processing and deep learning\r\n\r\n<b>Long-term research goal:<\/b> Deep learning methods have shown intriguing opportunities for natural language processing (NLP). Many typical tasks can be handled using end-to-end neural network\u2013based models. However, it\u2019s still a challenging problem to employ the correlations among different tasks and design a unified architecture to deal with different task simultaneously. So, my research work focuses on deep learning and multitask learning in NLP, hoping that a more general framework can be used to handle various tasks.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435555 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Xuan-Lu.jpg\" alt=\"Xuan Lu Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"http:\/\/sei.pku.edu.cn\/~luxuan13\/\" target=\"_blank\" rel=\"noopener\">Xuan Lu<\/a><\/h3>\r\nPeking University\r\n\r\nSupervisors: <a href=\"http:\/\/sei.pku.edu.cn\/~meih\/index_en.html\" target=\"_blank\" rel=\"noopener\">Hong Mei<\/a> and Xuanzhe Liu\r\n\r\n<b>Research interests:<\/b> Software analytics for mobile computing systems and applications\r\n\r\n<b>Long-term research goal:<\/b> The wide adoption of smartphones and tablet computers has triggered a surge of developing mobile applications (apps) in recent years. App stores, such as Apple Store and Google Play, continuously attract millions of developers to release and popularize their apps. At the same time, the competition among apps is increasingly fierce. In this context, developers are eager to know how users behave in apps and how they evaluate the app in app stores, so that they can better understand users and find problems in their apps to tackle. The long-term goal of my research is to provide directive suggestions to developers to optimize mobile apps based on user behavior analysis.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435552 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Chang-Hyun-Park.jpg\" alt=\"Chang Hyun Park Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3><a href=\"https:\/\/iamchanghyunpark.github.io\/\" target=\"_blank\" rel=\"noopener\">Chang Hyun Park<\/a><\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/calab.kaist.ac.kr\/~jhuh\/\" target=\"_blank\" rel=\"noopener\">Jaehyuk Huh<\/a>\r\n\r\n<b>Research interests:<\/b> Virtual memory and the underlying memory hierarchy\r\n\r\n<b>Long-term research goal:<\/b> My research interests are focused toward the computer architecture and system software stack of computing. Designing computer systems cannot be driven exclusively by either computer architects or system software developers. The two groups need to work hand-in-hand to design efficient systems. My research is focused on the memory systems of both hardware and software. My research interests and goals include improving virtual memory systems, which are increasing in how much they are relied on. My goal is to help prepare virtual memory systems for such a future.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435576 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Zhaofan-Qiu.jpg\" alt=\"Zhaofan Qiu Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Zhaofan Qiu<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisors: Xinmei Tian and Tao Mei\r\n\r\n<b>Research interests:<\/b> Deep learning, video understanding, and multimedia analysis\r\n\r\n<b>Long-term research goal:<\/b> My previous research mainly focused on building video understanding systems based on deep learning techniques. Following this direction, I will further explore the new possibilities of video understanding. The first dimension is to build an advanced video understanding system with capabilities such as action localization and action detection. Unlike traditional action recognition tasks, action localization and detection need to find when and where the action happens, which is much more challenging. The second dimension is to identify more powerful video representation approaches. One of the possible solutions is to adapt the image representation to the video domain, as well as training a very deep 3D CNN or deep RNN to improve the performance of current video understanding systems.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435567 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Yu-Wu.jpg\" alt=\"Yu Wu Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Yu Wu<\/h3>\r\nBeihang University\r\n\r\nSupervisors: Zhoujun Li and <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/mingzhou\/\" target=\"_blank\" rel=\"noopener\">Ming Zhou<\/a>\r\n\r\n<b>Research interests:<\/b> Dialog systems and natural language processing\r\n\r\n<b>Long-term research goal:<\/b> Chatbots are bringing us to a new technology era, the era of conversational interface. It is an era that won\u2019t require a mouse or a keyboard, where human and machine can communicate with each other through conversations. My research goal is to develop intelligent chatbots that are able to communicate with humans for various topics, using the large amount of real conversation data on the Internet. My current research focuses on context-aware chatbots, and achieves significant improvement in contrast to conventional approaches. In the future, my research will focus on two aspects: 1) Let chatbots interact with human proactively. 2) Integrate prior knowledge into chatbots.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-435573 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2017\/10\/Yuxiang-Yang.jpg\" alt=\"Yuxiang Yang Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Yuxiang Yang<\/h3>\r\nUniversity of Hong Kong\r\n\r\nSupervisor: <a href=\"http:\/\/qift.weebly.com\/giulio-chiribella.html\/\" target=\"_blank\" rel=\"noopener\">Giulio Chiribella<\/a>\r\n\r\n<b>Research interests:<\/b> Quantum information and computation\r\n\r\n<b>Long-term research goal:<\/b> The past few years witnessed both a rapid development of quantum technologies and new challenges in implementing long-distance quantum communication and long-term storage of quantum bits. My research focuses on implementing efficient quantum protocols and algorithms, which are tailor-made to reduce the consumption of quantum resources. I have been working on compression protocols for probes generated by quantum sensors and quantum algorithms with the minimum energy consumption. For the next step, I will focus on incorporating these protocols and algorithms in complex networks of quantum devices, aiming at achieving the quantum advantage with limited quantum resources. [\/panel] [panel header=\"2016 fellows\"]\r\n<h2>2016 fellows announced<\/h2>\r\nOver one hundred distinguished PhD candidates from 40 leading research academic institutions in China, Hong Kong SAR, Taiwan, Japan, Korea, and Singapore were nominated for the 2016 Microsoft Research Asia fellowships. After evaluating each application, our review committee selected 10 outstanding PhD candidates whose exceptional talent and innovation in computer science-related research identifies them as emerging leaders in the Asia-Pacific region.\r\n\r\n[caption id=\"attachment_318728\" align=\"aligncenter\" width=\"810\"]<img class=\"wp-image-318728\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/all-on-stage-3.jpg\" alt=\"Microsoft Research Asia fellows 2016\" width=\"810\" height=\"458\" \/> Microsoft Research Asia 2016 fellows with Peter Lee, CVP, Microsoft Research (far right); Hsiao-Wuen Hon, CVP, Microsoft Asia-Pacific R&amp;D Group, Microsoft Research Asia (second from the left), and Baining Guo, Assistant Managing Director of Microsoft Research Asia (far left).[\/caption]\r\n\r\n[caption id=\"attachment_318740\" align=\"aligncenter\" width=\"808\"]<img class=\"wp-image-318740\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/all-with-Adi-2.jpg\" alt=\"Microsoft Research Asia fellows 2016\" width=\"808\" height=\"456\" \/> After Award Ceremony, Microsoft Research Asia 2016 fellows had roundtable meeting with Adi Shamir, 2002 Turing Award Recipient.[\/caption]\r\n\r\n&nbsp;\r\n\r\n<img class=\"alignleft wp-image-318527 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Chuang-Gan-1-e1478654999848.jpg\" alt=\"Chuang Gan Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Chuang Gan<\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/itcs.tsinghua.edu.cn\/yao\/\" target=\"_blank\" rel=\"noopener noreferrer\">Andrew Chi-Chih Yao<\/a>\r\n\r\n<b>Research interests:<\/b> Deep Learning, Computer Vision, and Multimedia\r\n\r\n<b>Long-term research goal:<\/b> The increasing ubiquity of devices capable of capturing videos has led to an explosion in the amount of recorded video content. Instead of \u201ceyeballing\u201d the videos for potential useful information, it is desirable to develop automatic video analysis and understanding algorithms. My research focuses on deep learning and its application to large-scale video understanding, e.g. video action recognition, multimedia event detection, video captioning, etc. Developing new algorithms to teach machines to understand the video content is my long-term research goal.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"http:\/\/cvlab.postech.ac.kr\/~hyeonwoonoh\/\" target=\"_blank\" rel=\"noopener noreferrer\">Hyeonwoo Noh<\/a><\/h3>\r\n<a href=\"http:\/\/cvlab.postech.ac.kr\/~hyeonwoonoh\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318530 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Hyeonwoo-Noh-e1478655129277.jpg\" alt=\"Hyeonwoo Noh Microsoft Research Asia Fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nPohang University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/cvlab.postech.ac.kr\/~bhhan\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bohyung Han<\/a>\r\n\r\n<b>Research interests:<\/b> Visual understanding based on natural language\r\n\r\n<b>Long-term research goal:<\/b> My research objective is building visual understanding algorithm based on natural language. Contrary to defining different problem setting for different types of visual understanding, using natural language as an interface for a visual understanding system provides a unified testbed for various visual understanding abilities. Solving multiple visual understanding problems with single algorithm based on natural language based interface invokes various research questions such as executing appropriate understanding function for different language based queries or sharing common visual concepts across multiple understanding functions. I believe finding solutions to these challenges is key to building generally applicable visual understanding algorithms.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"https:\/\/sites.google.com\/site\/zhangjiemarina\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jie Zhang<\/a><\/h3>\r\n<a href=\"https:\/\/sites.google.com\/site\/zhangjiemarina\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318533 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Jie-Zhang-e1478655175391.jpg\" alt=\"Jie Zhang Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nPeking University\r\n\r\nSupervisor: <a href=\"http:\/\/sei.pku.edu.cn\/~zhanglu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Lu Zhang<\/a>\r\n\r\n<b>Research interests:<\/b> Software testing\r\n\r\n<b>Long-term research goal: <\/b>When ensuring software quality, traditionally developers\/testers explore program behaviors through test augmentation, either manually or with the aid of automatic test generation tools. However, the current test augmentation approaches have limitations, due to which many software systems are put into service with many behaviors unexplored, incurring serious threats to software quality. To complement the traditional approaches and better explore the program under test, I propose Variation Deduction, which is a new methodology that aims to explore more program behaviors through generating program variants instead of augmenting tests. In my future work, I plan to find valuable variation operators and deduction constraints, and make Variation Deduction applicable and valuable.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"http:\/\/junnishida.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jun Nishida<\/a><\/h3>\r\n<a href=\"http:\/\/junnishida.net\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318536 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Jun-Nishida-e1478655211107.jpg\" alt=\"Jun Nishida Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nUniversity of Tsukuba\r\n\r\nSupervisor: Kenji Suzuki\r\n\r\n<b>Research interests: <\/b>Augmenting Social and Embodied Experiences among People by Wearable Devices, Human-Computer Interaction, Biosignal Processing, Virtual Reality, Haptics\r\n\r\n<b>Long-term research goal: <\/b>My research goal is to develop and establish new techniques and style for interpersonal communication that facilitate social and embodied interactions among people. Especially, I focus on devices that allow people to share and reproduce one\u2019s sensory and kinesthetic experiences to achieve natural communication in rehabilitations and design process. We have been proposing 1) a wearable suit that transforms a wearer\u2019s embodiment into that of a child for architectural design procedure, and 2) a paired wearable device that can share muscle activity bi-directionally for assisting rehabilitations. The achievement contributes to the technology for augmented human and benefits from concrete applications which reveals new aspects of human behaviors.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"http:\/\/home.cse.ust.hk\/~lchenad\/\" target=\"_blank\" rel=\"noopener noreferrer\">Li Chen<\/a><\/h3>\r\n<a href=\"http:\/\/home.cse.ust.hk\/~lchenad\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318539 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Li-Chen-e1478655247357.jpg\" alt=\"Li Chen Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nThe Hong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/www.cse.ust.hk\/~kaichen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kai Chen<\/a>\r\n\r\n<b>Research interests:<\/b> Computing infrastructure &amp; Datacenter networks\r\n\r\n<b>Long-term research goal: <\/b>In this era of Big Data, both the volume of the data and the complexity of models to make sense of data are growing rapidly. However, speed of commodity processor and the limited memory size of commodity server force applications to scale out to large clusters. Most of the recent interesting applications, such as web search, recommendation systems, and deep networks, run on clusters of thousands of machines for both small companies and large enterprises. My long term research goal is accelerating application performance in this large-scale computing setting, especially the datacenter networks, as I believe this area presents challenging problems that can shape the future computing stack. My main research projects have been along this line: designing efficient networking for large-scale DC applications, with focus on flow scheduling for distributed and parallel computing systems.\r\n\r\n<hr \/>\r\n\r\n<a href=\"https:\/\/comp.nus.edu.sg\/~loiluu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Loi LUU<\/a><\/h3>\r\n<a href=\"https:\/\/comp.nus.edu.sg\/~loiluu\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318542 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Loi-LUU-e1478655282616.jpg\" alt=\"Loi LUU Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nNational University of Singapore\r\n\r\nSupervisor: <a href=\"https:\/\/comp.nus.edu.sg\/~prateeks\/\" target=\"_blank\" rel=\"noopener noreferrer\">Prateek Saxena<\/a>\r\n\r\n<b>Research interests:<\/b> Blockchain-based Cryptocurrencies and Distributed Consensus Algorithms\r\n\r\n<b>Long-term research goal: <\/b>Blockchain-based cryptocurrencies, such as Bitcoin and 250 similar alt-coins, embody at their core a blockchain protocol \u2014 a mechanism for a distributed network of computational nodes to periodically agree on a set of new transactions. Designing a secure blockchain protocol is very challenging because often the protocol is run in open networks with presence of adversary and without any trusted PKI, or nodes do not have inherent identities. My long-term research goal is to improve the security and scalability of these agreement protocols. To achieve this goal, my current researches focus on three main directions. The first direction is to study and understand the current design of existing blockchain systems like Bitcoin or Ethereum. The main and most important piece of my research is to design secure and scalable blockchain protocols, given all the knowledge and analyses from existing blockchain systems. The third and interesting research direction is to devise new use-cases for blockchains and improve the usability of blockchain protocols.\r\n\r\n<hr \/>\r\n\r\n<img class=\"alignleft wp-image-318545 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Qiang-Liu-e1478655356343.jpg\" alt=\"Qiang Liu Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/>\r\n<h3>Qiang Liu<\/h3>\r\nChinese Academy of Sciences\r\n\r\nSupervisor: <a href=\"http:\/\/nlpr-web.ia.ac.cn\/english\/irds\/People\/lwang\/lwang.html\" target=\"_blank\" rel=\"noopener noreferrer\">Liang Wang<\/a>\r\n\r\n<b>Research interests:<\/b> Data mining, user modeling, recommender systems, deep learning\r\n\r\n<b>Long-term research goal: <\/b>My research focuses on user modeling, i.e., how to describe a user based on his or her online behavioral logs and generated contents. This topic has two parts. The first part is normal user modeling, which utilizes users\u2019 daily data to analyze their interests and demands, and predict what they would like to do next. The long-term goal of this part of research is to model a variety of contextual information and multimodal contents in daily applications for learning better user representations. The second part is abnormal user modeling, which utilizes patterns in normal data to detect abnormal behaviors, suspicious users, as well as misinformation on the web for security propose. The long-term goal of this part of research is to learn representations from large-scale daily data on the web, and make unsupervised detection and judgement of abnormal data.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"http:\/\/ipads.se.sjtu.edu.cn\/doku.php?id=pub:members:wenhao_li\" target=\"_blank\" rel=\"noopener noreferrer\">Wenhao Li<\/a><\/h3>\r\n<a href=\"http:\/\/ipads.se.sjtu.edu.cn\/doku.php?id=pub:members:wenhao_li\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318548 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Wenhao-Li-e1478655368274.jpg\" alt=\"Wenhao Li Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nShanghai Jiao Tong University\r\n\r\nSupervisor: <a href=\"http:\/\/ipads.se.sjtu.edu.cn\/doku.php?id=pub:members:haibo_chen\" target=\"_blank\" rel=\"noopener noreferrer\">Haibo Chen<\/a>\r\n\r\n<b>Research interests:<\/b> OS and Architecture, System and Mobile Security\r\n\r\n<b>Long-term research goal: <\/b>My research goal is to bring systems with a good balance of performance, security and convenience to the real-world users. The approach I take is to improving the security and dependability of modern systems with practical hardware and software technologies. I am particularly interested in hardware-assisted security solutions which the industry is likely to adopt. I am currently exploring various forms of hardware-assisted and software designs in two aspects. The first is leveraging new hardware features (ARM TrustZone, Intel SGX, Intel VT, etc.) to improve the dependability and security of current mobile and cloud systems. The second is building and optimizing high performance platforms that benefit from system security.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"http:\/\/home.ustc.edu.cn\/~xiayingc\/\" target=\"_blank\" rel=\"noopener noreferrer\">Yingce Xia<\/a><\/h3>\r\n<a href=\"http:\/\/home.ustc.edu.cn\/~xiayingc\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318551 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Yingce-Xia-e1478655410780.jpg\" alt=\"Yingce Xia Microsoft Research Asia fellow\" width=\"170\" height=\"210\" \/><\/a>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisor: <a href=\"http:\/\/staff.ustc.edu.cn\/~ynh\/\" target=\"_blank\" rel=\"noopener noreferrer\">Nenghai Yu<\/a> and <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/tyliu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Tie-Yan Liu<\/a>\r\n\r\n<b>Research interests: <\/b>Machine Learning, Artificial Intelligence\r\n\r\n<b>Long-term research goal: <\/b>While deep learning is making surprising progress in past few years, large amounts of labeled samples are needed for the training procedure. However, human labeling is usually very costly, leading to limited training data scale. To overcome such the difficulty, in our previous work, we propose the dual learning framework that can efficiently leverage unlabeled data to overcome labeled data shortage. We take neural machine translation (NMT) as the first step and achieve significant improvements. Towards this end, my long-term research goal is to fulfill such a dual learning framework, which mainly includes two aspects: 1) Enrich dual learning with more techniques and further improve NMT performances based on the enriched framework; 2) Apply dual learning to other areas like speech recognition\/text to speech, image caption\/image generation and so on.\r\n\r\n<hr \/>\r\n\r\n<h3><a href=\"http:\/\/www4.comp.polyu.edu.hk\/~cszqcao\/\" target=\"_blank\" rel=\"noopener noreferrer\">Ziqiang Cao<\/a><\/h3>\r\n<a href=\"http:\/\/www4.comp.polyu.edu.hk\/~cszqcao\/\" target=\"_blank\" rel=\"noopener noreferrer\"><img class=\"alignleft wp-image-318554 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2016\/04\/Ziqiang-Cao-e1478655438806.jpg\" alt=\"Ziqiang Cao\" width=\"170\" height=\"210\" \/><\/a>\r\nThe Hong Kong Polytechnic University\r\n\r\nSupervisor: <a href=\"http:\/\/www4.comp.polyu.edu.hk\/~cswjli\/\" target=\"_blank\" rel=\"noopener noreferrer\">Wenjie Li<\/a>\r\n\r\n<b>Research interests:<\/b> Natural language processing, deep learning\r\n\r\n<b>Long-term research goal:<\/b> The application of deep learning is growing fashionable in natural language processing. However, the generated features seldom have the explicit meaning. Therefore, my research goal is to develop both effective and understandable techniques for natural language processing, especially automatic summarization. To this end, I plan to make full use of the traditional research outputs in natural language processing. I am interested in the combination of deep learning and prior knowledge to better bridge the gap between natural language and structured data. [\/panel] [panel header=\"2015 fellows\"]\r\n<h3>Wei Bai<\/h3>\r\nThe Hong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/www.cse.ust.hk\/~kaichen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kai Chen<\/a>\r\n\r\n<b>Research interests:<\/b> Data center networking\r\n\r\n<b>Long-term research goal:<\/b> Many data centers have been built around the world to provide various services and applications. In data centers, hundreds of thousands of servers are interconnected by using data center networks. My long-term research goal is to improve the performance of data center networks. To achieve this goal, I am focusing on designing new transport mechanisms to satisfy the low latency requirements of today\u2019s cloud applications. In the future, I plan to continue working on data center transport on two fronts: designing new solutions for multi-queue multi-service data centers as well as improving network performance for large-scale Remote Direct Memory Access deployments.\r\n\r\n<hr \/>\r\n\r\n<h3>Dong Deng<\/h3>\r\nTsinghua University\r\n\r\nSupervisors: <a href=\"http:\/\/dbgroup.cs.tsinghua.edu.cn\/fengjh\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jianhua Feng<\/a>, <a href=\"http:\/\/dbgroup.cs.tsinghua.edu.cn\/ligl\/\" target=\"_blank\" rel=\"noopener noreferrer\">Guoliang Li<\/a>\r\n\r\n<b>Research interests: <\/b>Database, data management, data cleaning, and data integration\r\n\r\n<b>Long-term research goal:<\/b> My research focuses on data cleaning, i.e., how to deal with errors and inconsistencies in information systems. As an example, in many applications such as data integration, commercial organizations need to collect data from various sources to conduct analysis and make decisions. The data from these different sources typically contain inconsistencies. For example, the same person\u2019s name may be misspelled or have different formats, such as with or without middle name. Such inconsistencies make it more challenging to link records from different places and answer queries approximately. My long-term research goal is to develop algorithms in order to make query answering and information retrieval efficient in the presence of such inconsistencies and errors.\r\n\r\n<hr \/>\r\n\r\n<h3>Li Dong<\/h3>\r\nBeihang University\r\n\r\nSupervisors: Wei Li, Ke Xu\r\n\r\n<b>Research interests: <\/b>Natural language processing, artificial intelligence\r\n\r\n<b>Long-term research goal:<\/b> My research goal is to develop novel and effective techniques for natural language processing, especially question ansering. In order to build better natural language interfaces for machines, we need to learn to sufficiently represent text from large volumes of data, and bridge the gap between natural language and structured data. I am interested in leveraging extensive unannotated data in a weak supervision setting, and building neural network-based frameworks to attack the problems of question answering.\r\n\r\n<hr \/>\r\n\r\n<h3>Hojoon Lee<\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/gsis.kaist.ac.kr\/cysec\/brentkang\/brentkang\" target=\"_blank\" rel=\"noopener noreferrer\">Brent Byunghoon Kang<\/a>\r\n\r\n<b>Research interests: <\/b>OS security, virtualization\r\n\r\n<b>Long-term research goal: <\/b>My research goal is to enhance the security and reliability of modern systems from the operating system level. I am particularly interested in building secure computing architectures by using trusted execution environments (TEEs). Modern computing systems are under severe threat: software vulnerabilities are constantly discovered and malwares are infecting billions. TEEs provide a safe execution environment in which system integrity monitoring or system transactions can be performed. I am currently exploring various forms of secure kernel design that involves TEE. My goal is to make fundamental changes to the traditional kernel architectures such that it becomes more resilient against software attacks.\r\n\r\n<hr \/>\r\n\r\n<h3>Tae-Hyun Oh<\/h3>\r\nKorea Advanced Institute of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/rcv.kaist.ac.kr\/\" target=\"_blank\" rel=\"noopener noreferrer\">In So Kweon<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision and machine learning (especially low-rank and sparse structure)\r\n\r\n<b>Long-term research goal: <\/b>My research interest is to investigate favorable non-convex models for tremendous applications. It has been known that global optimal solutions obtained from convex formulations may not often correspond to physically desirable solutions under many practical conditions. Since there is no limit of convexity while formulating non-convex models, flexible designs can be considered and the working range of algorithms can be broadened. In this stream, my future research will focus on: 1) Non-convex modeling that has favorable properties. It can be achieved by carefully looking into a problem and data. 2) Developing scalable optimizers corresponding to the newly developed formulations. I would like to make a stepping stone between engineering and theoretical people by providing useful and easy solutions.\r\n\r\n<hr \/>\r\n\r\n<h3>Yingwei Pan<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisor: <a href=\"http:\/\/staff.ustc.edu.cn\/~lihq\/index.html\" target=\"_blank\" rel=\"noopener noreferrer\">Houqiang Li<\/a>\r\n\r\n<b>Research interests: <\/b>Video understanding, large-scale visual search, click-through data analysis\r\n\r\n<b>Long-term research goal: <\/b>One of the fundamental problems in vision and multimedia is to bridge the semantic gap between visual content and text. In our previous research, we studied the problem from the viewpoint of cross-view (visual-semantic) embedding, that is, integrate content, structure, and\/or click data for learning a joint space that connects different natures of vision and text. Following the research in this direction, I will further explore the complex semantic relationship between long video sequence and natural language by considering three main challenges: 1) learning an effective representation of a long video, 2) video understanding from individual tags to natural sentence, and 3) demonstrating the success of our technologies in a wide range of applications such as video temporal localization and sentence (story) generation.\r\n\r\n<hr \/>\r\n\r\n<h3>Hao Tang<\/h3>\r\nPeking University\r\n\r\nSupervisor: <a href=\"http:\/\/sei.pku.edu.cn\/~zhanglu\/\" target=\"_blank\" rel=\"noopener noreferrer\">Lu Zhang<\/a>\r\n\r\n<b>Research interests: <\/b>Software analysis and testing\r\n\r\n<b>Long-term research goal: <\/b>Modern programming language features (e.g., callbacks, reflection, file I\/O) provide convenient language support for user-defined and platform-specific external behaviors. However, it is difficult to analyze unknown external behaviors with static analyzers. By now, we have made a small step towards analyzing these behaviors. I propose conditional reachability (formulated as tree-adjoining-language reachability) to handle callbacks when performing reachability analysis for software libraries. In my future work, I plan to improve the efficiency, scalability, precision, and generality of conditional reachability and apply it to real-life mobile and web applications. Also, I will continue to exploit new algorithms and data structures to analyze other types of unknown external behaviors.\r\n\r\n<hr \/>\r\n\r\n<h3>Tatsunori Taniai<\/h3>\r\nThe University of Tokyo\r\n\r\nSupervisor: <a href=\"http:\/\/www.hci.iis.u-tokyo.ac.jp\/~ysato\/\" target=\"_blank\" rel=\"noopener noreferrer\">Yoichi Sato<\/a>\r\n\r\n<b>Research interests: <\/b>Mathematical optimization for low- and mid-level computer vision\r\n\r\n<b>Long-term research goal: <\/b>I am interested in applications of mathematical optimization techniques for low- and mid-level computer vision problems, such as 3D shape recovery from images, dense correspondence estimation, and image segmentation. My long-term research goal is to establish a joint framework of image co-segmentation and dense correspondence estimation where common object regions between input images are precisely segmented and aligned as output. Such rich information about object regions and positions is useful in many applications such as fine-grained object detection, scene parsing, and image editing.\r\n\r\n<hr \/>\r\n\r\n<h3>Hao Wang<\/h3>\r\nThe Hong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"https:\/\/www.cse.ust.hk\/~dyyeung\/\" target=\"_blank\" rel=\"noopener noreferrer\">Dit-Yan Yeung<\/a>\r\n\r\n<b>Research interests: <\/b>Machine learning, data mining, Bayesian deep learning, social network analysis, recommender systems\r\n\r\n<b>Long-term research goal: <\/b>There are two primary classes of tasks in AI. The first class is perception tasks like visual object recognition and text understanding. These are the tasks that can be easily completed by normal human beings. The second class is inference and planning. This involves a higher level of intelligence like decision making, data analysis, and logic deduction. In a sense, deep learning with its multiple processing layers is better at the first class of tasks and probabilistic graphical models with its Bayesian nature excel at the second class of tasks. The problem is that in the real world, these two classes often entangle with each other. Our research goal is to combine the power of deep learning and Bayesian models in a single principled framework to get the best of both worlds. We call this framework Bayesian Deep Learning. So far, we have successfully applied Bayesian Deep Learning for applications like recommendation, relational learning, and link prediction. In the future, we would like to extend our work both horizontally, to handle more applications, and vertically, to provide more theoretical guarantees along with more sophisticated models.\r\n\r\n<hr \/>\r\n\r\n<h3>Feng-Long Xie<\/h3>\r\nHarbin Institute of Technology\r\n\r\nSupervisors: <a href=\"http:\/\/homepage.hit.edu.cn\/pages\/lihaifeng\/\" target=\"_blank\" rel=\"noopener noreferrer\">Haifeng Li<\/a>, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/frankkps\/\" target=\"_blank\" rel=\"noopener noreferrer\">Frank K. Soong<\/a>\r\n\r\n<b>Research interests: <\/b>Cross-lingual speech synthesis, voice conversion, speaker adaptation\r\n\r\n<b>Long-term research goal: <\/b>Cross-lingual TTS (text-to-speech) synthesis can be defined as synthesizing speech in the target language (L2) with a specific speaker\u2019s recorded speech in his native language (L1) and to maintain this speaker\u2019s voice characteristics. Currently, I am trying to use Speaker Independent Deep Neural Networks to equalize the speaker difference and Kullback-Leilbler divergence to measure the acoustic units\u2019 difference. In the future, I will apply this framework to many speech applications, e.g., voice conversion, speaker adaptation, speech recognition, language learning, and talking head.\r\n\r\n<hr \/>\r\n\r\n<h3>Ting Zhang<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisors: Yong Wang, <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/hon\/\" target=\"_blank\" rel=\"noopener noreferrer\">Hsiao-Wuen Hon<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision, machine learning\r\n\r\n<b>Long-term research goal: <\/b>My research interests have revolved around computer vision and machine learning, with a specific focus on fast approximate nearest neighbor (ANN) search. The goal of fast ANN search is to find the closest item of the given query accurately as well the fastest of all of the database vectors. To achieve that goal, our previous work uses composite quantization to represent the database vectors by compact codes. My long-term research goal is to continue exploring algorithms to improve the search accuracy as well as the search efficiency of the approximate nearest neighbors. I believe that many applications\u2014such as similar image search, object retrieval, and so on\u2014can benefit from fast ANN search.\r\n\r\n<hr \/>\r\n\r\n<h3>Yongfeng Zhang<\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/www.tsinghua.edu.cn\/publish\/csen\/4623\/2010\/20101224180015486255888\/20101224180015486255888_.html\" target=\"_blank\" rel=\"noopener noreferrer\">Shaoping Ma<\/a>\r\n\r\n<b>Research interests: <\/b>Personalization theories, computational economics, sentiment analysis\r\n\r\n<b>Long-term research goal: <\/b>I believe that academic research should advance itself not only for commercial benefits, but also for the benefit of society. Economists, philosophers, and sociologists devote their lives for a better human society in our physical world, and we as computer scientists should also push the virtual online society towards a more fair and just world. The research that I am targeting now and in the future falls into the emerging research field of computational economics. I especially care about the new frameworks for online services such as e-commerce, online financing, or freelancing based on the maximization of the joint surplus of both service consumers and providers (i.e., the social surplus) so that online services are matched from providers to consumers in a way that benefits the virtual society for a social good.\r\n\r\n<hr \/>\r\n\r\n<h3>Zhenzhe Zheng<\/h3>\r\nShanghai Jiao Tong University\r\n\r\nSupervisor: <a href=\"http:\/\/cs.nju.edu.cn\/gchen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Guihai Chen<\/a>\r\n\r\n<b>Research interests: <\/b>Algorithmic network economics, wireless network, mobile computing, and cloud computing\r\n\r\n<b>Long-term research goal: <\/b>Generally, I am interested in research problems that lie at the interaction of economic and computer science. My recent research interests mainly focus on resource management in computer networks. With the rapid development of distributed systems, networking systems, and cloud computing technology, modern network resource management needs the cooperation of different network entities. However, the goals of selfish and rationale network entities usually conflict with the overall goal of the whole network system. My research goal is to apply the methodology of game theory to design efficient mechanisms for network resource management in different layers of network systems. I am also interested in designing efficient and economic-robust market systems, such as online ad auctions and data markets. [\/panel] [panel header=\"2014 fellows\"]\r\n<h3>Szu-Wen Fan<\/h3>\r\nKeio University\r\n\r\nSupervisor: Masahiko Inami\r\n\r\n<b>Research interests: <\/b>Substitutional reality (SR), human computer interaction (HCI), cross-modal perception, human's perception of reality\r\n\r\n<b>Long-term research goal: <\/b>I am enthusiastic about finding out how humans perceive reality, based on the multi-modal stimulation we receive with our five senses. As we are well aware, our beliefs about reality are limited to what we perceive with our sensory organs, which could be misdirected from true reality. Therefore, we can immerse users in alternate reality experiences by just stimulating authentic sensory feedbacks using a computer. However, these alternate reality experiences are limited to one reality only. Users are in either actual reality or alternate reality. I suggest that we can extend a human\u2019s perception of reality to be composed of more than one reality simultaneously, by focusing on cross-modal perception and seamless transition between actual and alternate reality. I am looking to blend realities:\r\n<ol>\r\n \t<li>From remote reality, for immersive telepresence<\/li>\r\n \t<li>From past reality, for immersive memories<\/li>\r\n \t<li>From wearable augmented sensory reality, for augmenting human ability<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Seunghoon Hong<\/h3>\r\nPohang University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/cv.postech.ac.kr\/~bhhan\" target=\"_blank\" rel=\"noopener noreferrer\">Bohyung Han<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision and machine learning\r\n\r\n<b>Long-term research goal: <\/b>I am interested in developing a visual tracking system that can reliably track the target over \u201clong-term\u201d videos. Although substantial progress has been made in the tracking literature, this problem still remains challenging since most existing trackers are susceptible to model drift and temporary tracking failures. To resolve such issues, my previous research has mainly focused on developing reliable graphical models for tracking. By identifying a suitable order of tracking within a video, more reliable tracking can be achieved robust to intermediate failures. My future research would focus on long-term visual tracking based on high-level context understanding. Many real-world videos, such as movies and TV shows, contain drastic variations of the target and scene, and tracking based only on low-level visual information would inevitably fail with these challenges. I believe that contexture information, such as scene structure, object categories, and so forth, provides additional information to resolve ambiguities in tracking and to find more meaningful structure of the video for tracking.\r\n\r\n<hr \/>\r\n\r\n<h3>Wenping Hu<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisors: Yong Wang, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/hon\/\" target=\"_blank\" rel=\"noopener noreferrer\">Hsiao-Wuen Hon<\/a>\r\n\r\n<b>Research interests: <\/b>Computer-aided language learning, speaker recognition, speech recognition, deep learning\r\n\r\n<b>Long-term research goal: <\/b>Computer-Aided Language Learning (CALL) systems, powered by the advancement of speech technology, can bridge the gap between the supply and demand of language teachers and have become ubiquitous learning tools with widely used mobile devices. We have built our own CALL system and transferred it into Microsoft Bing dictionary. This system is able to evaluate language learners\u2019 pronunciation quality, automatic detect the mispronunciation error within the word, and give some constructive feedbacks to language learners, e.g., common error patterns. Currently, I am using deep learning for pronunciation proficiency assessment and deficiency detection. We trained our golden pronunciation model by deep neural networks, investigated more robust pronunciation algorithms, applied transfer-learning for pronunciation deficiency detection and speeded up the real-time system to make it practical. In the future, we will further extend our algorithms for free-style, spontaneous speech assessment, which will have an assessment of learner's spoken pronunciation proficiency based on his or her free talk recordings, just like TOEFL oral test. Without given the read transcription, it will be more challenging, but more helpful and practical for second language learners.\r\n\r\n<hr \/>\r\n\r\n<h3>Yanyan Jiang<\/h3>\r\nNanjing University\r\n\r\nSupervisor: Jian Lu\r\n\r\n<b>Research interests: <\/b>Testing and analysis of concurrent programs\r\n\r\n<b>Long-term research goal: <\/b>Software quality assurance is becoming a major challenge for the modern software industry, as software products are getting larger, more complex, and parallel. Software testing practice, being the most prevalent approach for improving software quality, is also encountering emerging challenges. I am particularly interested in testing and analysis techniques to improve the quality of concurrent programs. Recently, researchers have proposed deterministic replay, symbolic analysis, and predictive analysis. These techniques points out a new trend for testing real-world concurrent software. I am conducting my research towards the following goals:\r\n<ol>\r\n \t<li>Record and replay a concurrent program execution that can last days long within affordable resource consumption<\/li>\r\n \t<li>Design sound and semantic-aware causal execution model for concurrent programs that can be checked in polynomial time<\/li>\r\n \t<li>Using symbolic analysis and constraint solving technique to systematically explore interleaving space of concurrent programs<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Hiroshi Kajino<\/h3>\r\nThe University of Tokyo\r\n\r\nSupervisors: <a href=\"http:\/\/www.ibis.t.u-tokyo.ac.jp\/yamanishi\/index_e.html\" target=\"_blank\" rel=\"noopener noreferrer\">Kenji Yamanishi<\/a>, <a href=\"http:\/\/www.geocities.co.jp\/Technopolis\/5893\/index_e.html\" target=\"_blank\" rel=\"noopener noreferrer\">Hisashi Kashima<\/a>\r\n\r\n<b>Research interests: <\/b>Human computation, crowdsourcing, machine learning, data mining\r\n\r\n<b>Long-term research goal: <\/b>My research goal is to establish a secure and reliable infrastructure for human computation. Human computation is an emerging computer science technique in which machines utilize humans as computational resources to tackle computationally hard problems. Crowdsourcing is often used to allow machines to access human resources via the web. However, it has been often pointed out that crowdsourcing brings about a reliability issue that a result of computation is sometimes unreliable. My research focuses not only on the reliability issue, but also on privacy issues that arise, for example, when a task is submitted to crowdsourcing and when workers in crowdsourcing return computation results. By addressing the privacy issues, the applicability of human computation will become wider than ever.\r\n\r\n<hr \/>\r\n\r\n<h3>Yuan Lin<\/h3>\r\nFudan University\r\n\r\nSupervisor: <a href=\"http:\/\/homepage.fudan.edu.cn\/zhangzz\/en\/\" target=\"_blank\" rel=\"noopener noreferrer\">Zhongzhi Zhang<\/a>\r\n\r\n<b>Research interests: <\/b>Random walks in complex network\r\n\r\n<b>Long-term research goal: <\/b>Random walks are a fascinating research topic attracting intensive attention within the scientific community, not only for their intrinsic mathematical beauty but also for their important implications to various scientific issues, such as target searching, node ranking, computer vision, to name just a few. My research goal will focus on the following two tasks.\r\n<ol>\r\n \t<li>Discovering the behavior of biased random walks in complex networks. Although unbiased random walks have been widely investigated, many dynamics should be described by biased random walks, which is still an outstanding issue. My work will both numerically and analytically research the influence of different biases on the crucial random walk indicators.<\/li>\r\n \t<li>Applying new random walk mechanisms in practical problems. Based on the obtained exhibitions of various biased random walks, my research will explore the applications of such bias in virus spreading, link prediction, community detection, and so on.<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Yutao Liu<\/h3>\r\nShanghai Jiao Tong University\r\n\r\nSupervisor: <a href=\"http:\/\/ipads.se.sjtu.edu.cn\/haibo_chen\" target=\"_blank\" rel=\"noopener noreferrer\">Haibo Chen<\/a>\r\n\r\n<b>Research interests: <\/b>Cloud and mobile security\r\n\r\n<b>Long-term research goal: <\/b>My primary research goal is to improve server and mobile security in system perspective. My current research projects investigate two areas of focus:\r\n<ol>\r\n \t<li>Leveraging potential architecture support to improve the dependability and security of current systems, specifically, using hardware features in different architectures to protect data privacy, enforce control flow integrity, actively monitor the whole system, timely detect potential attack, and so forth.<\/li>\r\n \t<li>Building and optimizing reliable virtualization environment to benefit system security. Virtualization has been a tremendous success, especially in server consolidation and security enhancement. However, there are still some problems when considering security aspects, and my goal is to ameliorate virtualization architecture to adapt to increasing security requirements.<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Canyi Lu<\/h3>\r\nNational University of Singapore\r\n\r\nSupervisors: <a href=\"http:\/\/www.ece.nus.edu.sg\/stfpage\/eleyans\/\" target=\"_blank\" rel=\"noopener noreferrer\">Shuicheng Yan<\/a>, <a href=\"http:\/\/www.cis.pku.edu.cn\/faculty\/vision\/zlin\/zlin.htm\" target=\"_blank\" rel=\"noopener noreferrer\">Zhouchen Lin<\/a>\r\n\r\n<b>Research interests: <\/b>Compressive sensing of block diagonal affinity matrix, convex optimization\r\n\r\n<b>Long-term research goal: <\/b>I am interested in developing algorithms for important problems with support in theory. In particular, I am aiming to establish two or three representative works in my PhD career. They include:\r\n<ol>\r\n \t<li>Compressive sensing of block diagonal affinity matrix Affinity matrix is almost everywhere. The ideal affinity matrix should be block diagonal, i.e., data points from different subspaces or subjects are not similar at all. There are many known methods, such as sparse representation and low-rank representation, which lead to the block diagonal solution under certain conditions. We have established the Enforced Block Diagonal (EBD) conditions to unify many previous methods. We are aiming to find a convex model for block diagonal affinity matrix pursuit and to establish the theory for the exact recovery.<\/li>\r\n \t<li>Alternating Direction Method of Multiplier (ADMM) ADMM is the most important solver for convex optimization with linear constraint. It regains much attention in recent years due to the success of compressive sensing. There are many ADMM variants based on different structures of the objective function, but their convergence proofs are given case by case. We are aiming to give a general framework that unifies many known ADMMs with a unified proof. Also, there are still many open problems with ADMM. I aim to give a comprehensive study in the future.<\/li>\r\n<\/ol>\r\nI am also interested in applying our algorithms and theory for computer vision and pattern recognition.\r\n\r\n<hr \/>\r\n\r\n<h3>Yongtae Park<\/h3>\r\nKorea University\r\n\r\nSupervisor: Hyogon Kim\r\n\r\n<b>Research interests: <\/b>Software defined radio for mobile devices, vehicular networking\r\n\r\n<b>Long-term research goal: <\/b>I am working on Software Defined Radio (SDR) for mobile devices. Recently, I demonstrated the feasibility of ZigBee communication on smartphones and am working to show the possibility of supporting various protocols without using dedicated chips. Providing these protocols as an app will catalyze the development and maintenance of protocols on future mobile devices. Also, it will enable people to develop the personalized and secure protocols for themselves. To make this happen, I am working on various issues such as optimization of SDR processing, fairness of antenna sharing, and so forth. Ultimately, my long-term research goal is to make mobile devices speak thousands of protocols. I hope my research plays an important role to facilitate \u201cInternet of Things\u201d by helping to connect humans with ambient smart devices.\r\n\r\n<hr \/>\r\n\r\n<h3>Shaoqing Ren<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisors: <a href=\"http:\/\/staff.ustc.edu.cn\/~binli\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bin Li<\/a>, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/jiansun\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jian Sun<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision, detection, and localization for face and generic objects\r\n\r\n<b>Long-term research goal: <\/b>My research interest lies in computer vision, especially detection and localization. Detection and localization as the fundamental problems in computer vision have been studied widely and are closing to practical applications. My previous research focuses on highly efficient detection and localization (alignment) on human face and generic objects, which makes these computer vision technologies cheaper on computation cost and closer to everybody\u2019s daily life. I envision more and more applications will benefit from the breakthrough of image understanding, especially apps on mobile devices. My future research will continue to explore accurate and highly efficient detection\/localization systems and make them more practical for real applications.\r\n\r\n<hr \/>\r\n\r\n<h3>Yingxia Shao<\/h3>\r\nPeking University\r\n\r\nSupervisor: <a href=\"http:\/\/net.pku.edu.cn\/~cuibin\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bin Cui<\/a>\r\n\r\n<b>Research interests: <\/b>Large-scale graph analysis, parallel computing framework, scalable data processing\r\n\r\n<b>Long-term research goal: <\/b>With the rapid growth of social networks, e-commence, and sensor networks, large-scale graphs become easily available in public. These graphs represent the complex relationships in the real world and contain a lot of valuable knowledge. Therefore, mining and analyzing these large-scale graphs is an important approach to discover the knowledge. The long-term goal of my research is to scale graph analysis to the scales of real-world graphs and develop a scalable and efficient large graph analysis toolkit. To achieve this goal, we design new graph algorithms according to several principles. First, the new algorithms will exploit the new characteristics of graphs, such as sparse, dynamic, and power-law distribution. Second, the workload patterns of graph problems will be unraveled as well. Lastly, for intricate graph metrics (e.g., SimRank, closeness, betweenness, and so forth), we will design new approximation algorithms. The toolkit will be carefully designed by abstracting primitive operations among above graph analysis task.\r\n\r\n<hr \/>\r\n\r\n<h3>Shuai Yi<\/h3>\r\nThe Chinese University of Hong Kong\r\n\r\nSupervisor: <a href=\"http:\/\/www.ee.cuhk.edu.hk\/~xgwang\/\" target=\"_blank\" rel=\"noopener noreferrer\">Xiaogang Wang<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision and crowd video surveillance\r\n\r\n<b>Long-term research goal: <\/b>I am now focusing on crowd video analysis in computer vision. Nowadays, with steady population growth and worldwide urbanization, crowded situations are more common. Crowd management and traffic control are common problems in public areas with a high population density. Millions of surveillance cameras are capturing these areas day and night. However, it is still very challenging to analyze these videos accurately and efficiently. A pedestrian\u2019s decision-making is very complex and many factors may have great influence on one\u2019s walking behavior. For example, the structure of the scene, the inner belief of the pedestrian, and the interaction with other moving pedestrians and stationary groups will all change pedestrians\u2019 walking paths and make the problem difficult to solve. I am now focusing my analysis on human behaviors in crowds and trying to find the relationship between a pedestrian\u2019s walking behavior and corresponding influence factors. In the future, I hope a joint pedestrian decision making model can be built which will aid in behavior analysis and abnormal detection. [\/panel] [panel header=\"2013 fellows\"]\r\n<h3>Chen Cao<\/h3>\r\nZhejiang University\r\n\r\nSupervisor: <a href=\"http:\/\/www.kunzhou.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kun Zhou<\/a>\r\n\r\n<b>Research interests: <\/b>Computer graphics, real-time facial animation\r\n\r\n<b>Long-term research goal: <\/b>My research goal is to propose a real-time facial animation system for average users. Several issues need to be addressed to implement such a system. Primarily, the usage of special equipment such as markers or camera arrays should be avoided. Alternatively, an ordinary web camera should be the only capture device. Our system expectations are to be used in a wide range of environments, including both indoor and outdoor environments exposed to direct sunlight. The system should be capable of handling the changes in lighting and background. It is also vital to ensure maximum high performance system capabilities with the system. I will continue to focus on researching real-time facial animation in the future. There are many problems that need to be addressed and fixed with Real-time Facial Animation. Our current system requires a large amount of training data during the preprocessing step. It is ideal to find a solution to reduce the training data requirements, because the user needs to perform 60 different facial expressions and head poses. A key goal is to implement our real-time facial animation system on mobile devices. This, however, is not straightforward. Due to the limited computational power on mobile devices, the three-dimensional shape regression is time consuming. The limitations in our current system to handle dramatic lighting changes must be addressed since mobile devices are frequently used in various environments.\r\n\r\n<hr \/>\r\n\r\n<h3>Dong Chen<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisors: <a href=\"http:\/\/staff.ustc.edu.cn\/~binli\/\" target=\"_blank\" rel=\"noopener noreferrer\">Bin Li<\/a>, <a class=\"invalidLink\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/jiansun\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jian Sun<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision, face recognition, face detection\r\n\r\n<b>Long-term research goal: <\/b>Face recognition has been widely studied for more than three decades. Facial recognition of an image is divided into four steps: face detection, face alignment, feature extraction, and feature learning. Each element will have an impact on the overall performance. My long-term research goal is to improve face recognition accuracy during the four-step process. The following implementations will highlight my long-term research goal.\r\n<ol>\r\n \t<li>A more efficient structure for face detection and the face alignment pipeline will be designed. Face recognition\u2019s speed and accuracy, and the Landmark\u2019s Localization Algorithm will be improved.<\/li>\r\n \t<li>An illumination invariant feature combined with data-driven based approaches to increase face recognition accuracy under uncontrolled extreme lighting conditions will be designed.<\/li>\r\n \t<li>I will extend my feature and learning method to more face-related tasks such as attribute recognition, as well as exploiting these technologies to further promote the recognition rate.<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Luwei Cheng<\/h3>\r\nThe University of Hong Kong\r\n\r\nSupervisors: <a href=\"http:\/\/www.cs.hku.hk\/~clwang\" target=\"_blank\" rel=\"noopener noreferrer\">Cho-Li Wang<\/a>, <a href=\"http:\/\/www.cs.hku.hk\/~fcmlau\" target=\"_blank\" rel=\"noopener noreferrer\">Francis Chi Moon Lau<\/a>\r\n\r\n<b>Research interests: <\/b>Virtualization technologies for cloud computing\r\n\r\n<b>Long-term research goal: <\/b>Virtualization is largely adopted in datacenters to support on-demand cloud computing. Current virtualization technologies are primarily invented to host unmodified or slightly modified operating systems using virtual machines (VMs). However, the existence of the hypervisor, an additional layer between the hardware and the guest operating system has significantly changed the assumptions of many protocols, which are originally designed for physical running environments. In particular, VM scheduling delays due to CPU sharing can negatively affect many kernel services such as the process scheduler, the disk scheduler and network protocols. Consequently, my research efforts go towards deeply integrating the hypervisor and the guest OS to form a more lightweight container for upper-layer applications.\r\n\r\n<hr \/>\r\n\r\n<h3>Quan Fang<\/h3>\r\nChinese Academy of Sciences\r\n\r\nSupervisor: <a href=\"http:\/\/nlpr-web.ia.ac.cn\/mmc\/homepage\/csxu.html\" target=\"_blank\" rel=\"noopener noreferrer\">Changsheng Xu<\/a>\r\n\r\n<b>Research interests: <\/b>Georeferenced social media research and application\r\n\r\n<b>Long-term research goal: <\/b>My research interest lies in georeferenced social media research and application. I focus on geo-social computing in particular, which aims to sense and mine the massive georeferenced social media data to understand the user and geo-location for better serving. I am working towards the following strategies. Georeferenced media mining and modeling aims to sense and mine the massive geo-tagged data to understand geo-location, especially from the human-sensed perspective. User modeling in social media is to understand users by exploiting rich online user-generated content. I will combine and model the user, geo-location, and content in a unified framework. I am dedicated to developing efficient and effective techniques for harvesting knowledge about users and geo-locations from georeferenced social media data. The derived knowledge can benefit various user-centric applications.\r\n\r\n<hr \/>\r\n\r\n<h3>Ping Luo<\/h3>\r\nThe Chinese University of Hong Kong\r\n\r\nSupervisor: <a href=\"http:\/\/www.ie.cuhk.edu.hk\/people\/xotang.shtml\" target=\"_blank\" rel=\"noopener noreferrer\">Xiaoou Tang<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision, computer graphics, and machine learning\r\n\r\n<b>Long-term research goal: <\/b>I am interested in the areas of computer vision, computer graphics, and machine learning. My particular focus is deep learning and its applications with vision and graphics problems, including face and human detection, parsing (poses, clothing), attribute analysis, and identification. I envision that information technology in the future can help automatically find, identify, and connect people all over the world from various media\u2014such as photographs, social networking, and images on the Internet\u2014making everybody\u2019s life easier. My future research heads toward this goal both theoretically and practically. Image representation is a fundamental problem in computer vision. Although this representation can be directly learned from data by deep learning, how to learn or fuse the representations from different domains is still a great challenge. I will first devise a cross-domain multi-task method for image modeling so that it can fuse the data from different sources, while serving as the building block for many high-level tasks; such as detection, recognition, and segmentation. High computational cost is another challenge when working with large amounts of data. My second task will focus on building a large-scale distributed deep-learning framework, which can effectively deal with tens of millions of image data.\r\n\r\n<hr \/>\r\n\r\n<h3>Masayasu Ogata<\/h3>\r\nKeio University\r\n\r\nSupervisor: <a title=\"\" href=\"http:\/\/www.ayu.ics.keio.ac.jp\/~michita\/\" target=\"_blank\" rel=\"noopener noreferrer\">Michita Imai<\/a>\r\n\r\n<b>Research interests: <\/b>Human-computer interaction, natural user interface\r\n\r\n<b>Long-term research goal: <\/b>I am interested in the area of HCI, especially natural user interface and the embodied interaction with sensing technique design and development, as well as the wearable device and interactive system. Our experiences with the living environment are getting complicated with user motivation and desired functionality. To ease and illuminate interfaces in daily life, I will propose a vision and an idea to increase interface transparency that makes the user unaware of the computer\u2019s existence when they access the everyday environment. Sensing techniques and designing appropriate feedback to the user are required for such an achievement. My experience and skill will reinforce the long-term goal that integrates the human and the system in a daily environment in the near future.\r\n\r\n<hr \/>\r\n\r\n<h3>Jianping Shi<\/h3>\r\nThe Chinese University of Hong Kong\r\n\r\nSupervisor: <a href=\"http:\/\/www.cse.cuhk.edu.hk\/~leojia\/\" target=\"_blank\" rel=\"noopener noreferrer\">Jiaya Jia<\/a>\r\n\r\n<b>Research interests: <\/b>Computer vision, machine learning\r\n\r\n<b>Long-term research goal: <\/b>I have an extensive interest in computer vision and machine learning. My previous research focuses mainly on the model\u2019s point of view, including Sparse and Low-rank Matrix Factorization. These models have been widely used in many applications. To be proficient in modeling is not enough. I believe building appropriate models to fit the vision data is much more important. I am currently working on several projects, including Image Blur Analysis, Human Detection, and RGBD Video Processing. Such projects necessitate a deep understanding of the image data. For example, the Blur Analysis is based on several natural image statistics. Our Human Co-detection System explores the intrinsic similarity for the same person across photos. The RGBD Visual Tracking utilizes the depth distribution of the tracking object. All these projects stem from vision data understanding. My future research will continue to explore computer vision\u2019s interesting aspects, while also advancing the computer to understand our colorful world.\r\n\r\n<hr \/>\r\n\r\n<h3>Shusen Wang<\/h3>\r\nZhejiang University\r\n\r\nSupervisor: <a href=\"http:\/\/bcmi.sjtu.edu.cn\/~zhzhang\/\" target=\"_blank\" rel=\"noopener noreferrer\">Zhihua Zhang<\/a>\r\n\r\n<b>Research interests: <\/b>Machine learning, matrix analysis, convex optimization\r\n\r\n<b>Long-term research goal: <\/b>My research focus is on enabling large-scale machine learning using randomized approximations. As we know, many classical machine learning methods suffer from high time and space complexities and are thus prohibitive in big-data applications. Some randomized approximation approaches\u2014for example, random projection, random selection, and the Nystrom method\u2014reduce the time and space complexities of many machine-learning methods from quadric or cubic to near linear by sacrificing a little accuracy. I am working on improving the existing approximation methods, devising new approximation methods, and providing theoretical guarantees for the methods. My goal is to make the randomized approximation approaches more accurate and more efficiently computed.\r\n\r\n<hr \/>\r\n\r\n<h3>Huanjing Yue<\/h3>\r\nTianjin University\r\n\r\nSupervisors: Feng Wu, <a href=\"http:\/\/www.tju.edu.cn\/seie\/english\/fas\/faculty\/201106\/t20110605_105661.html\" target=\"_blank\" rel=\"noopener noreferrer\">Jingyu Yang<\/a>\r\n\r\n<b>Research interests: <\/b>Cloud-based image coding and processing\r\n\r\n<b>Long-term research goal: <\/b>My long-term research goal is to advance the state-of-the-art technologies in signal processing and computer vision through cloud data. The number of images shared on the Internet is dramatically increasing due to the dramatic increase in social websites. This provides a live and huge image base, which has posed demands for the development and implementation of content-based image retrieval. In addition, it brings new opportunities for many well-known, ill-posed image processing and computer vision problems, such as image composition, completion, and colorization. I mainly focus on cloud-based image coding, super-resolution, and denoising. Different from the traditional methods, our work proposes utilizing the external correlated images to improve coding efficiency and image restoration quality. Our work is an early step toward a new era of cloud-based image coding and processing.\r\n\r\n<hr \/>\r\n\r\n<h3>Jun Zhang<\/h3>\r\nNanyang Technological University\r\n\r\nSupervisor: <a href=\"http:\/\/www3.ntu.edu.sg\/home\/xkxiao\/\" target=\"_blank\" rel=\"noopener noreferrer\">Xiaokui Xiao<\/a>\r\n\r\n<b>Research interests: <\/b>Database management and data privacy\r\n\r\n<b>Long-term research goal: <\/b>There is a growing emphasis in the world at large on utilizing and disseminating aggregate statistics from demographic data, health records, Internet activity, and other sources. This data, however, can seldom be accessed for public studies due to concerns over individual privacy. Existing methods for privacy preservation still offers inadequate assurance, because they rely on restrictive assumptions on how a malicious user may attack the data, whereas those assumptions can be easily violated in practice. Motivated by the practical importance of privacy protection and the deficiencies of the existing methods, my research work aims to conduct a comprehensive study on mathematically rigorous privacy protection models. Specifically, I am interested in:\r\n<ol>\r\n \t<li>Generic and fundamental algorithms under Differential Privacy, a state-of-the-art privacy model that is rigorously formulated based on statistics, offering an extremely strong privacy guarantee;<\/li>\r\n \t<li>Novel class of privacy definitions with better trade-off between privacy and utility.<\/li>\r\n<\/ol>\r\n[\/panel] [panel header=\"2012 fellows\"]\r\n<h3>Yang Cao<\/h3>\r\nBeihang University\r\n\r\nSupervisor: Jinpeng Huai\r\n\r\n<b>Research interests:<\/b> Database theory and systems, graph pattern matching, social searching\r\n\r\n<b>Long-term research goal:<\/b>\r\n<ol>\r\n \t<li>Graph pattern matching revised for social searching: Graph pattern matching is fundamental to social analysis. Traditional techniques like subgraph isomor-phism and graph simulation either impose too strong a topological constraint on graphs to retrieve meaningful matches and incur high computational complexity (for example, subgraph isomorphism is NP-complete), or are too loose to find correct matches. Moreover, data graphs for modeling social networks are usually evolving over time. Existing models and techniques often fall short of handling this without incurring high computational complexity. To this end, I will propose new graph pattern matching models (queries) as well as techniques (algorithms) for evaluating them. The model should be well-designed such that it could strike a balance between the express power (semantics) and the computational complexity for evaluating them, and should be able to handle the evolvement of data graphs as well.<\/li>\r\n \t<li>Keywords search on graph data: Keywords searching has been a longstanding issue for search engines. Traditionally, keyword searching is often conducted on structured and semistructured data, as well as documents. Existing models and techniques for keywords searching cannot be naively adopted for social search engines. Recent research about keywords searching on graph data mainly focuses on developing techniques for retrieving matched results on graph data. My research for this aspect will instead turn to proposing new keywords searching models specific for graph data, to take the advantages of graph pattern queries to provide more accurate semantics for emerging applications.<\/li>\r\n \t<li>New computational model and complexity classes: In the context of large-scale data, traditional computational classes are too generic for identifying problems that are solvable practically, and moreover, they cannot take properties of data graphs (social networks) into consideration. To this end, I will propose a computational model and a hierarchy of complexity classes, to provide a dichotomy between those queries that are feasible on large-scale data graphs and those that are not, and thus properly classify social searching problems, with regard to the express power and evaluating complexity of the queries.<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Menglei Chai<\/h3>\r\nZhejiang University\r\n\r\nSupervisor: <a href=\"http:\/\/kunzhou.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kun Zhou<\/a>\r\n\r\n<b>Research interests:<\/b> Computer Graphics, Computer Vision, Image Processing, and so forth\r\n\r\n<b>Long-term research goal: <\/b>My long-term research goal primarily targets portrait image manipulation, especially for recovering semantic information from common images and providing a series of techniques which enables useful applications for average users. As we know, images and videos featuring humans as their subjects are of great interest for both industrial experts and average end users, and have motivated various kinds of research in computer graphics during the past decades. But few of them involve 3-D information to resolve the occlusions and ambiguities issues, making these low-level image processing methods of limited application for specific tasks of portrait manipulation. Our research work will focus on portrait image manipulation, especially for components such as face, hair, body shape, clothing, and so forth. The key thought is to extract 3-D information by utilizing specific prior knowledge achieved with machine-learning techniques, and apply them to drive a broad range of user-friendly applications which were previously challenging.\r\n\r\n<hr \/>\r\n\r\n<h3>Jun Kato<\/h3>\r\nThe University of Tokyo\r\n\r\nSupervisor: <a href=\"http:\/\/www-ui.is.s.u-tokyo.ac.jp\/~takeo\/\" target=\"_blank\" rel=\"noopener noreferrer\">Takeo Igarashi<\/a>\r\n\r\n<b>Research interests:<\/b> Human-computer interaction and programming language (user interface for programmers)\r\n\r\n<b>Long-term research goal: <\/b>I am interested in the broad area of human-computer interaction, but have been especially focused on interactions between humans and the real world through computers (physical computing and robot applications), their input modalities (natural user interfaces), and their development methods (prototyping toolkits and integrated development environments). In my future vision, everyone can be a weekend carpenter who makes his\/her everyday life easy and comfortable with help of information technology\u2014in other words, end-user programming of the real world in the real world. Currently, I am planning to take two steps toward this goal.\r\n<ul>\r\n \t<li>First, I will propose development environments that improve the programmer\u2019s experience (Programmer\u2019s eXperience, PX) on \u201creal-world programming\u201d that involves interactions with the real world, such as robot control and vision-based object recognition.<\/li>\r\n \t<li>Second, I will propose end-user development environments for the same \u201creal-world programming\u201d that make creation of customized applications feasible to end users through specialized user interfaces.<\/li>\r\n<\/ul>\r\nGiven my past experience on PX research, I expect that further investigation on the professional user interfaces will serve as the fundamentals for designing new user interfaces for the end users.\r\n\r\n<hr \/>\r\n\r\n<h3>Yin-Hsi Kuo<\/h3>\r\nNational Taiwan University\r\n\r\nSupervisor: <a href=\"http:\/\/www.csie.ntu.edu.tw\/~winston\/\" target=\"_blank\" rel=\"noopener noreferrer\">Winston H. Hsu<\/a>\r\n\r\n<b>Research interests:<\/b> Multimedia content analysis, image retrieval, and mobile visual search\r\n\r\n<b>Long-term research goal: <\/b>Nowadays, the rapid computational power of mobile devices brings the emerging need for mobile visual search. Different from the traditional content-based retrieval system, the mobile devices have the ability to process the query (for example, feature extraction) before the transmission. In recent years, the hash-based method has become promising for approximate nearest neighbor (ANN) search because of its ability to deal with high-dimensional features and large-scale databases in an efficient way. An efficient and effective method on the mobile devices is extremely crucial; hence, the hash-based approach becomes a possible direction in reality. In the future, we attempt to integrate the hash-based approach with contextual information to achieve efficient and scalable mobile visual search.\r\n\r\n<hr \/>\r\n\r\n<h3>Hongjin Liang<\/h3>\r\nUniversity of Science and Technology of China\r\n\r\nSupervisor: <a href=\"http:\/\/staff.ustc.edu.cn\/~xyfeng\/\" target=\"_blank\" rel=\"noopener noreferrer\">Xinyu Feng<\/a>\r\n\r\n<b>Research interests:<\/b> Program verification\r\n\r\n<b>Long-term research goal: <\/b>The computer industry has shown explosive growth in the past years. This trend is not likely to slow down. However, the lack of reliable and secure software is becoming a bottleneck for such growth. In particular, programming on multiprocessors is extremely error-prone, but very difficult to debug. Thus, in my long-term research career, I would like to develop new and easy-to-use technologies and tools that could improve the reliability, safety, and security of software on multiprocessors. I am particularly interested in:\r\n<ol>\r\n \t<li>Designing program logics for verifying the correctness (including safety and liveness properties) of concurrent programs<\/li>\r\n \t<li>Developing automatic or semi-automatic tools to support concurrent program verification<\/li>\r\n \t<li>Applying my technologies and tools to verify today\u2019s concurrent libraries (such as java.concurrent.util), runtime support in concurrent systems (such as concurrent garbage collectors), and other concurrent software used in practice<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Yongxin Tong<\/h3>\r\nHong Kong University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/www.cs.ust.hk\/~leichen\/\" target=\"_blank\" rel=\"noopener noreferrer\">Lei Chen<\/a>\r\n\r\n<b>Research interests:<\/b> Database and data mining, machine learning\r\n\r\n<b>Long-term research goal: <\/b>My research interests mainly focus on uncertain data management and mining. With the emergence of many real applications\u2014such as sensor network monitoring, moving object search, protein-protein interaction (PPI) network analysis, and so forth\u2014managing and mining uncertain data has attracted much attention in the database and data mining communities recently. Although current researches have solved some fundamental operations over uncertain data\u2014for example, join, ranking, mining frequent item sets, clustering, and so forth\u2014there are a few works to explore the hidden correlation in uncertain data due to the intricate probabilistic structure and high computational complexity. Therefore, my research aims to address these challenges via incorporating both theoretical and practical viewpoints.\r\n<ul>\r\n \t<li>On the theoretical side, I will first design a novel model to capture intrinsic correlated properties in uncertain data. Additionally, I will propose a series of efficient and effective algorithms in order to adapt complex structural data, such as uncertain data streams and uncertain graphs. This is particularly important in the age of big data as well.<\/li>\r\n \t<li>On the practical side, I hope to extend our theoretical model and algorithms to real application scenarios, in other words, I try to develop a new crowdsourcing platform based on mobile computing and utilize our probabilistic model to handle the uncertainty control in this system, which is one of most urgent problems in the crowdsourcing researches.<\/li>\r\n<\/ul>\r\nIn summary, my research goal is to better discover and manage the hidden correlation and rules over massive uncertain data effectively.\r\n\r\n<hr \/>\r\n\r\n<h3>Xinggang Wang<\/h3>\r\nHuazhong University of Science and Technology\r\n\r\nSupervisor: <a href=\"http:\/\/mc.eistar.net\/MCWebDisplay\/PersonDetails.aspx?Name=Wenyu%20Liu\" target=\"_blank\" rel=\"noopener noreferrer\">Wenyu Liu<\/a>\r\n\r\n<b>Research interests:<\/b> Computer vision and machine learning\r\n\r\n<b>Long-term research goal: <\/b>My research interests are computer vision and machine learning, especially the problem of object detection. Object detection concerning detecting instances of semantic objects of a certain class in digital images and videos, is a fundamental problem in computer vision, and has wide applications in people\u2019s lives, for example, face recognition, image search, automatic driving, and so forth. Through years of research, the problem of frontal face detection has been well studied, and this technology has been integrated into products. However, detecting most of the other objects in the real world is a very difficult problem due to the great variations of object appearances, such as persons and cars. My previous research mainly focuses on part-based object detection, shape-based object detection, and semi-supervised learning for object detection. It brings together shape feature design, object model design, machine learning, and optimization. Based on my previous research, my long-term research goal is to build an object detection system in a very weakly supervised way, which can achieve the state-of-the-art performance and be used in some real-life applications. I will put more effort on:\r\n<ol>\r\n \t<li>Studying discriminative object part detector; for example, people can easily recognize a leopard when looking at dapple of leopard. Part detector is robust to occlusion. Ensemble of part detector can reach good object detector.<\/li>\r\n \t<li>Using context information for object detection which combines scene understanding and object recognition.<\/li>\r\n \t<li>Combining modeling based methods with data-driven based methods.<\/li>\r\n<\/ol>\r\n\r\n<hr \/>\r\n\r\n<h3>Yizhong Zhang<\/h3>\r\nZhejiang University\r\n\r\nSupervisor: <a href=\"http:\/\/kunzhou.net\/\" target=\"_blank\" rel=\"noopener noreferrer\">Kun Zhou<\/a>\r\n\r\n<b>Research interests:<\/b> Computer graphics, physically based simulation\r\n\r\n<b>Long-term research goal: <\/b>My research interest is in physically based simulation, especially simulation of complex natural phenomena for graphics. Physically based simulation is a powerful tool in both science and engineering. It can help us to predict unknown phenomena that may prevent the success of an experiment or the improvement of a product's design before manufacturing. In graphics, we need simulations to make animations of water, fire, cloth, and many other phenomena that contain high frequency details. We want the simulation to be fast while being reliable so that it will be easier for artists to control the animation. My research goal is to improve the performance of simulation, including propose reduced physical model, method of efficient and robust handling of geometry, and parallel computing. By using these methods, we can get high quality simulation results more quickly.\r\n\r\n<hr \/>\r\n\r\n<h3>Xin Zhao<\/h3>\r\nPeking University\r\n\r\nSupervisor: <a href=\"http:\/\/net.pku.edu.cn\/~lxm\/\" target=\"_blank\" rel=\"noopener noreferrer\">Li Xiaoming<\/a>\r\n\r\n<b>Research interests:<\/b> Social media analysis, web text mining, machine learning\r\n\r\n<b>Long-term research goal: <\/b>My general research interests are within the area of online social networks(for example,Twitter and Facebook) analysis. Online social networks\u2014in particular, their rapid growth and development\u2014are continuously attracting users all around the world, which significantly changes the way that people live. Although various text mining methods have been shown effective to deal with traditional document collections, for example, scientific publications, very few of them are tested to achieve very robust and sound performance on real data sets of online social networks due to the fact we still do not fully understand the nature of online social networks, such as their underlying information structure, user behaviors, and connecting patterns. To overcome these difficulties, the goal of my research is to develop both principled methodologies and innovative applications for automatically analyzing and discovering knowledge from online social networks. Specifically, I will mainly focus on the aspect of content analysis of online social networks in the long term. More challenging than traditional content analysis, we have to first understand the underlying information patterns and uncover the generative process of such information before we can construct effective models. Previous research experiences will be helpful for me to identify and solve real-world problems that are valuable to common users in this direction. During the problem-solving process, I will try to construct formal methods with clear and intuitive motivations, borrow the ideas and techniques from multiple disciplines, and evaluate research results with large-scale real data.\r\n\r\n<hr \/>\r\n\r\n<h3>Chong Zu<\/h3>\r\nTsinghua University\r\n\r\nSupervisor: <a href=\"http:\/\/itcs.tsinghua.edu.cn\/yao\" target=\"_blank\" rel=\"noopener noreferrer\">Chi-Chih Yao<\/a>\r\n\r\n<b>Research interests:<\/b> Quantum information processing\r\n\r\n<b>Long-term research goal: <\/b>Quantum information and computation is an interdisciplinary subject of computer science and quantum physics. During recent decades, there have been a lot of important theoretical results in this field, including unconditional secure cryptography, true random number generation, exponential speedup algorithm for factoring, and so on. However, realizing quantum information processing and quantum computation is still not an easy task. In my research, I am mainly focusing on two different physical systems, photonic qubit and NV center in diamond, both of which are promising candidates for future quantum information processing and computation.\r\n<ol>\r\n \t<li>Photonic qubit: In my recent research, we are working on a project to theoretically propose and experimentally build a brand-new, reliable, and practical quantum random number generator. By saying reliable, it means that we can always use some methods to certify that the random number we are producing is indeed from quantum power, rather than some classical device or lack knowledge of the device, and the randomness (characterized by min-entropy) can be bounded; while saying practical means that our random number generator can be efficient and easy to be realized. In the future, we plan to generate quantum state with more qubits, with which we can demonstrate interesting quantum information protocols, as well as realize large-scale quantum information processing.<\/li>\r\n \t<li>NV center in diamond: Recently, we have started to build NV center system. We want to use it to realize room-temperature quantum memory, as well as solid state quantum repeater for long-range quantum information. Our long-term goal is to build a hybrid quantum computation and information network with different quantum systems including trapped ion, matter qubit, and photonic qubit, which can be treated as a prototype for a future genuine quantum network.<\/li>\r\n<\/ol>\r\n[\/panel] [panel header=\"2011 fellows\"]\r\n<ul>\r\n \t<li><b>Dongzhe Ma<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Peiran Ren<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Xiaohui Bei<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Xun Cao<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Quan Wang<\/b>, Peking University<\/li>\r\n \t<li><b>Xiang Song<\/b>, Fudan University<\/li>\r\n \t<li><b>Chenglin Li<\/b>, Shanghai Jiao Tong University<\/li>\r\n \t<li><b>Guo Li<\/b>, Zhejiang University<\/li>\r\n \t<li><b>Shengjun Huang<\/b>, Nanjing University<\/li>\r\n \t<li><b>Jing Yuan<\/b>, University of Science and Technology of China<\/li>\r\n \t<li><b>Cuiling Lan<\/b>, Xidian University<\/li>\r\n \t<li><b>Guangmin Wang<\/b>, Xi\u2019an Jiao Tong University<\/li>\r\n \t<li><b>Yang Cao<\/b>, University of Science and Technology of Huazhong<\/li>\r\n \t<li><b>Qiang Hao<\/b>, Tianjin University<\/li>\r\n \t<li><b>Zhen Liao<\/b>, Nankai University<\/li>\r\n \t<li><b>Lei Cui<\/b>, Harbin Institute Technology University<\/li>\r\n \t<li><b>Xiaoshuai Sun<\/b>, Harbin Institute Technology University<\/li>\r\n \t<li><b>Bolei Zhou<\/b>, The Chinese University of Hong Kong<\/li>\r\n \t<li><b>Lin Ma<\/b>, The Chinese University of Hong Kong<\/li>\r\n \t<li><b>Nobuyuki Umetani<\/b>, The University of Tokyo<\/li>\r\n \t<li><b>Yefeng Liu<\/b>, Waseda University<\/li>\r\n \t<li><b>Hyunson Seo<\/b>, Yonsei University<\/li>\r\n \t<li><b>Yohan Chon<\/b>, Yonsei University<\/li>\r\n \t<li><b>Jaesik Park<\/b>, Korea Advanced Institute of Science and Technology<\/li>\r\n \t<li><b>Sangwon Seo<\/b>, Korea Advanced Institute of Science and Technology<\/li>\r\n \t<li><b>Kazi Rubaiat Habib<\/b>, National University of Singapore<\/li>\r\n \t<li><b>Gang Yu<\/b>, Nanyang Technological University<\/li>\r\n \t<li><b>Nikolay Gravin<\/b>, Nanyang Technological University<\/li>\r\n \t<li><b>Shuo-Hung Chen<\/b>, National Tsing Hua University<\/li>\r\n<\/ul>\r\n[\/panel] [panel header=\"2010 fellows\"]\r\n<ul>\r\n \t<li><b>Haris Javaid<\/b>, The University of New South Wales<\/li>\r\n \t<li><b>Yue Deng<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Chongyang Ma<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Xiong Li<\/b>, Shanghai Jiao Tong University<\/li>\r\n \t<li><b>Bo Geng<\/b>, Peking University<\/li>\r\n \t<li><b>Shiliang Zhang<\/b>, Institute of Computing Technology, The Chinese Academy of Science<\/li>\r\n \t<li><b>Xiulian Peng<\/b>, University of Science and Technology of China<\/li>\r\n \t<li><b>Xiao Zhang<\/b>, Tsinghua University<\/li>\r\n \t<li><b>Xinying Song<\/b>, Harbin Institute Technology University<\/li>\r\n \t<li><b>Jinbao Wang<\/b>, Harbin Institute Technology University<\/li>\r\n \t<li><b>Wei Wu<\/b>, Peking University<\/li>\r\n \t<li><b>Linghe Kong<\/b>, Shanghai Jiao Tong University<\/li>\r\n \t<li><b>Liang Wang<\/b>, Nanjing University<\/li>\r\n \t<li><b>Ping Chen<\/b>, Nanjing University<\/li>\r\n \t<li><b>Weiwei Wu<\/b>, University of Science and Technology of China<\/li>\r\n \t<li><b>Xiaojun Qian<\/b>, The Chinese University of Hong Kong<\/li>\r\n \t<li><b>Dongxiao Yu<\/b>, The University of Hong Kong<\/li>\r\n \t<li><b>Lu Wang<\/b>, The Hong Kong University of Science &amp; Technology<\/li>\r\n \t<li><b>Seokhwan Kim<\/b>, University of Tsukuba<\/li>\r\n \t<li><b>Adiyan Mujibiya<\/b>, The University of Tokyo<\/li>\r\n \t<li><b>Seungjin Lee<\/b>, Korea Advanced Institute of Science and Technology<\/li>\r\n \t<li><b>Tae-Joon Kim<\/b>, Korea Advanced Institute of Science and Technology<\/li>\r\n \t<li><b>Gae-won You<\/b>, Pohang University of Science and Technology<\/li>\r\n \t<li><b>Sungjin Lee<\/b>, Seoul National University<\/li>\r\n \t<li><b>Shenghua Gao<\/b>, Nanyang Technological University<\/li>\r\n \t<li><b>Yi-Ling Hsieh<\/b>, National Cheng Kung University<\/li>\r\n \t<li><b>Cheng-Te Li<\/b>, National Taiwan University<\/li>\r\n \t<li><b>Tsung-Te Lai<\/b>, National Taiwan University<\/li>\r\n<\/ul>\r\n[\/panel] [panel header=\"2009 fellows\"]\r\n<ul>\r\n \t<li><b>Novi Quadrianto<\/b>,<b> <\/b>The Australian National University<\/li>\r\n \t<li><b>Ashnil Kumar<\/b>,<b> <\/b>The University of Sydney<\/li>\r\n \t<li><b>William Voorsluys<\/b>,<b> <\/b>The University of Melbourne<\/li>\r\n \t<li><b>Shenghua Liu<\/b>,<b> <\/b>Tsinghua University<\/li>\r\n \t<li><b>Hao Wen<\/b>,<b> <\/b>Tsinghua University<\/li>\r\n \t<li><b>Zhiwei Xiong<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\r\n \t<li><b>Dong Liu<\/b>,<b> <\/b>Harbin Institute of Technology<\/li>\r\n \t<li><b>Bo Yu<\/b>,<b> <\/b>Harbin Institute of Technology<\/li>\r\n \t<li><b>Litian Tao<\/b>,<b> <\/b>Beihang University<\/li>\r\n \t<li><b>Yufeng Li<\/b>,<b> <\/b>Nanjing University<\/li>\r\n \t<li><b>Wei Wang<\/b>,<b> <\/b>Nanjing University<\/li>\r\n \t<li><b>Huanhuan Cao<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\r\n \t<li><b>Xiaoyin Wang<\/b>,<b> <\/b>Peking University<\/li>\r\n \t<li><b>Jun Lang<\/b>,<b> <\/b>Beijing Institute of Technology<\/li>\r\n \t<li><b>Derek Hao Hu<\/b>,<b> <\/b>The Hong Kong University of Science &amp; Technology<\/li>\r\n \t<li><b>Kaiming He<\/b>,<b> <\/b>The Chinese University of Hong Kong<\/li>\r\n \t<li><b>Tasuku Oonishi<\/b>, Tokyo Institute of Technology<\/li>\r\n \t<li><b>Yoshida Yuichi<\/b>, Kyoto University<\/li>\r\n \t<li><b>Jun Hatori<\/b>, The University of Tokyo<\/li>\r\n \t<li><b>Jongwuk Lee<\/b>, Pohang University of Science and Technology<\/li>\r\n \t<li><b>Jung-Tae Lee<\/b>,<b> <\/b>Korea University<\/li>\r\n \t<li><b>Bingjun Zhang<\/b>,<b> <\/b>National University of Singapore<\/li>\r\n \t<li><b>Lixin Duan<\/b>,<b> <\/b>Nanyang Technological University<\/li>\r\n \t<li><b>Yu-Chen Sun<\/b>,<b> <\/b>National Chiao Tung University<\/li>\r\n \t<li><b>Kai-yin Chen<\/b>,<b> <\/b>National Taiwan University<\/li>\r\n<\/ul>\r\n[\/panel] [panel header=\"2008 fellows\"]\r\n<ul>\r\n \t<li><b>Feng Zhang<\/b>,<b> <\/b>Beijing Institute of Technology<\/li>\r\n \t<li><b>Xiangyi Meng<\/b>,<b> <\/b>Beijing Institute of Technology<\/li>\r\n \t<li><b>Tong Wu<\/b>,<b> <\/b>Beijing University of Post and Telecommunication<\/li>\r\n \t<li><b>Li Xu<\/b>,<b> <\/b>Chinese University of Hong Kong<\/li>\r\n \t<li><b>Jun Lang<\/b>,<b> <\/b>Harbin Institute of Technology<\/li>\r\n \t<li><b>Qingqing Zhang<\/b>,<b> <\/b>Institute of Acoustics, Chinese Academy of Sciences<\/li>\r\n \t<li><b>Yanyan Lan<\/b>,<b> <\/b>Institute of Applied Mathematics, Academy of Mathematics and Systems Science, Chinese Academy of Sciences<\/li>\r\n \t<li><b>Xiaoqin Zhang<\/b>,<b> <\/b>Institute of Automation, Chinese Academy of Sciences<\/li>\r\n \t<li><b>Xiubo Geng<\/b>,<b> <\/b>Institute of Computing Technology, Chinese Academy of Sciences<\/li>\r\n \t<li><b>Souneil Park<\/b>,<b> <\/b>Korea Advanced Institute of Science and Technology<\/li>\r\n \t<li><b>Ki-woong Park<\/b>,<b> <\/b>Korea Advanced Institute of Science and Technology<\/li>\r\n \t<li><b>Dafan Dong<\/b>,<b> <\/b>Nankai University<\/li>\r\n \t<li><b>Yi Huang<\/b>,<b> <\/b>Nanyang Technological University<\/li>\r\n \t<li><b>Yu-Lin Wang<\/b>,<b> <\/b>National Cheng Kung University<\/li>\r\n \t<li><b>Yi-Hsuan Yang<\/b>,<b> <\/b>National Taiwan University<\/li>\r\n \t<li><b>Yantao Zheng<\/b>,<b> <\/b>National University of Singapore<\/li>\r\n \t<li><b>Lijiang Chen<\/b>,<b> <\/b>Peking University<\/li>\r\n \t<li><b>Sunghyun Cho<\/b>,<b> <\/b>Pohang University of Science and Technology<\/li>\r\n \t<li><b>YongDeok Kim<\/b>,<b> <\/b>Pohang University of Science and Technology<\/li>\r\n \t<li><b>Dikan Xing<\/b>,<b> <\/b>Shanghai Jiao Tong University<\/li>\r\n \t<li><b>Jingjing Fu<\/b>,<b> <\/b>The Hong Kong University of Science and Technology<\/li>\r\n \t<li><b>Jun Hong<\/b>,<b> <\/b>The University of Hong Kong<\/li>\r\n \t<li><b>Florian Mueller<\/b>,<b> <\/b>The University of Melbourne<\/li>\r\n \t<li><b>Tomoaki Higo<\/b>,<b> <\/b>The University of Tokyo<\/li>\r\n \t<li><b>Pinyan Lu<\/b>,<b> <\/b>Tsinghua University<\/li>\r\n \t<li><b>Jialin Zhang<\/b>,<b> <\/b>Tsinghua University<\/li>\r\n \t<li><b>Qiming Hou<\/b>,<b> <\/b>Tsinghua University<\/li>\r\n \t<li><b>Kun Xu<\/b>,<b> <\/b>Tsinghua University<\/li>\r\n \t<li><b>SYifei Don<\/b>,<b> <\/b>University of New South Wales<\/li>\r\n \t<li><b>Hao Xu<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\r\n \t<li><b>Yuan Liu<\/b>,<b> <\/b>University of Science and Technology of China<\/li>\r\n \t<li><b>Myung - Suk Song<\/b>,<b> <\/b>Yonsei University<\/li>\r\n<\/ul>\r\n[\/panel] [\/accordion]"},{"id":4,"name":"Testimonials","content":"<h2>The fellowship experience: in their own words<\/h2>\r\nFellows talk about their experiences before, during, and after their fellowship tenure.\r\n\r\n<a href=\"https:\/\/youtu.be\/NdVH36ONjDA\" target=\"_blank\" rel=\"noopener\">Watch the video &gt;<\/a>\r\n<h3>Opened my mind about research and careers<\/h3>\r\n<blockquote>\u201cI am very fortunate to have a relationship with Microsoft Research Asia, which I've always admired as one of the best research laboratories in the world. I jumped into MSRA as an intern during my master's studies. Every day\u2019s discussions with my mentor were insightful, and I have developed a sense of what makes 'good' research and built courage to pursue challenging research in PhD. The Fellowship award, which back then looked too competitive for me, has given me confidence and opened my mind to many career possibilities.\u201d \u2014 <strong>Tatsunori Taniai<\/strong>, University of Tokyo, 2015 Fellow<\/blockquote>\r\n<h3>Alumni community for academic exchange<\/h3>\r\n<blockquote>\"During the application, both my PhD advisor and the head of department were very supportive and encouraging about my fellowship. The award aspects of the Microsoft Research Asia Fellowship that attract me most are the opportunity to do an internship in Microsoft Research Asia and to join an alumni community for academic exchange.\" \u2014 <strong>Yingwei Pan<\/strong>, University of Science and Technology of China, 2015 Fellow<\/blockquote>\r\n<h3>Develop confidence as a researcher<\/h3>\r\n<blockquote>\"The Microsoft Research Asia Fellowship award has helped me develop confidence as a researcher and provides me encouragement to do high-quality research work as well as responsibility. The award gives me freedom to pursue the research topics that I am really interested in. It allows to get many opportunities to present and discuss with diverse and excellent researchers.\" \u2014<strong>Tae-Hyun Oh<\/strong>, Korea Advanced Institute of Science and Technology, 2015 Fellow<\/blockquote>\r\n<h3>Strong motivation for my research<\/h3>\r\n<blockquote>\"It is a great pleasure to receive a Microsoft Research Asia Fellowship. This prize has given me strong motivation for my research. My research is about multimedia retrieval, and I know that Microsoft Research Asia is also focusing on the same area.\" \u2014 <strong>Yin-Hsi Kuo<\/strong>, National Taiwan University, 2012 Fellow<\/blockquote>\r\n<h3>Relentlessly pursue higher learning<\/h3>\r\n<blockquote>\"I am especially grateful for the internship opportunity that comes with the Fellowship award. My internship period has been very exciting and I have gotten to work with many excellent people. They not only taught me computer-related knowledge and skills, but also team spirit, problem-solving approaches, and how to relentlessly pursue higher learning. It has also been eye-opening to talk to fellow research students with different research interests.\" \u2014 <strong>Canyi Lu<\/strong>, National University of Singapore, 2014 Fellow<\/blockquote>\r\n<h3>Collaborate with some of the best researchers in the world<\/h3>\r\n<blockquote>\"I think the Microsoft Research Asia Fellowship is a fantastic opportunity for young researchers like me to gain hands-on research experience in a large corporation. I get to collaborate with some of the best researchers in the world and the experience has helped me a lot in determining my own research direction. At the same time, the generous financial support provided by the Fellowship makes my PhD study a lot easier and my life more trouble-free.\" \u2014 <strong>Wei Bai<\/strong>, Hong Kong University of Science and Technology, 2015 Fellow<\/blockquote>"}],"msr_impact_theme":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-academic-program\/215593"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-academic-program"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-academic-program"}],"version-history":[{"count":46,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-academic-program\/215593\/revisions"}],"predecessor-version":[{"id":969618,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-academic-program\/215593\/revisions\/969618"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/786595"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=215593"}],"wp:term":[{"taxonomy":"msr-opportunity-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-opportunity-type?post=215593"},{"taxonomy":"msr-region","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-region?post=215593"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=215593"},{"taxonomy":"msr-program-audience","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-program-audience?post=215593"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=215593"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=215593"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}