{"id":724366,"date":"2021-02-24T12:37:50","date_gmt":"2021-02-24T20:37:50","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?p=724366"},"modified":"2022-11-01T09:00:05","modified_gmt":"2022-11-01T16:00:05","slug":"aaai-2021-accelerating-the-impact-of-artificial-intelligence","status":"publish","type":"post","link":"https:\/\/www.microsoft.com\/en-us\/research\/blog\/aaai-2021-accelerating-the-impact-of-artificial-intelligence\/","title":{"rendered":"AAAI 2021: Accelerating the impact of artificial intelligence"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-scaled.jpg\" alt=\"Graphic shows blue background with hexagons that each have a icon related to technology concepts i.e. cloud, binary code, eye, microchip, etc. \"\/><\/figure>\n\n\n<aside id=accordion-ec3c1ee8-cd47-43af-abcf-2a06a5fde1c3 class=\"msr-table-of-contents-block accordion mb-5 pb-0\" data-bi-aN=\"table-of-contents\">\n\t<button class=\"btn btn-collapse bg-gray-100 mb-0 display-flex justify-content-between\" type=\"button\" data-mount=\"collapse\" data-target=\"#accordion-collapse-ec3c1ee8-cd47-43af-abcf-2a06a5fde1c3\" aria-expanded=\"true\" aria-controls=\"accordion-collapse-ec3c1ee8-cd47-43af-abcf-2a06a5fde1c3\">\n\t\t<span class=\"msr-table-of-contents-block__label subtitle\">In this article<\/span>\n\t\t<span class=\"msr-table-of-contents-block__current mr-4 text-gray-600 font-weight-normal\" aria-hidden=\"true\"><\/span>\n\t<\/button>\n\t<div id=\"accordion-collapse-ec3c1ee8-cd47-43af-abcf-2a06a5fde1c3\" class=\"msr-table-of-contents-block__collapse-wrapper collapse show\" data-parent=\"#accordion-ec3c1ee8-cd47-43af-abcf-2a06a5fde1c3\">\n\t\t<div class=\"accordion-body bg-gray-100 border-top pt-4\">\n\t\t\t<ol class=\"msr-table-of-contents-block__list\">\n\t\t\t\t\t\t\t\t\t<li class=\"msr-table-of-contents-block__list-item\">\n\t\t\t\t\t\t<a href=\"#policy-optimization-using-reinforcement-learning-for-universal-trading\" class=\"msr-table-of-contents-block__list-item-link\">Policy optimization using reinforcement learning for universal trading<\/a>\n\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<li class=\"msr-table-of-contents-block__list-item\">\n\t\t\t\t\t\t<a href=\"#uwspeech-a-speech-to-speech-translation-system-that-doesnt-rely-on-written-text\" class=\"msr-table-of-contents-block__list-item-link\">UWSpeech: A speech-to-speech translation system that doesn\u2019t rely on written text<\/a>\n\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<li class=\"msr-table-of-contents-block__list-item\">\n\t\t\t\t\t\t<a href=\"#investigating-how-data-augmentation-affects-privacy-in-deep-learning-models\" class=\"msr-table-of-contents-block__list-item-link\">Investigating how data augmentation affects privacy in deep learning models<\/a>\n\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<li class=\"msr-table-of-contents-block__list-item\">\n\t\t\t\t\t\t<a href=\"#modeling-speech-and-noise-simultaneously-for-state-of-the-art-speech-enhancement\" class=\"msr-table-of-contents-block__list-item-link\">Modeling speech and noise simultaneously for state-of-the-art speech enhancement<\/a>\n\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t<\/ul>\n\t\t<\/div>\n\t<\/div>\n\t<span class=\"msr-table-of-contents-block__progress-bar\"><\/span>\n<\/aside>\n\n\n\n<p>The purpose&nbsp;of&nbsp;the&nbsp;Association for the Advancement of Artificial&nbsp;Intelligence, according to its bylaws, is&nbsp;twofold.&nbsp;The first is to promote research&nbsp;in the area of&nbsp;AI, and the second is to promote the&nbsp;<em>responsible<\/em>&nbsp;<em>use&nbsp;<\/em>of these types of technology.&nbsp;The&nbsp;result was a&nbsp;<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aaai.org\/Conferences\/AAAI-21\/\" target=\"_blank\" rel=\"noreferrer noopener\">35th AAAI Conference on Artificial Intelligence (AAAI-21)<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>&nbsp;schedule that&nbsp;broadens the possibilities&nbsp;of&nbsp;AI and&nbsp;is&nbsp;heavily reflective of&nbsp;a pivotal time in AI research&nbsp;when&nbsp;experts are asking bigger questions about how best to&nbsp;responsibly&nbsp;develop, deploy, and integrate&nbsp;the technology.&nbsp;&nbsp;<\/p>\n\n\n\n<div class=\"annotations \" data-bi-aN=\"citation\">\n\t<ul class=\"annotations__list card depth-16 bg-body p-4 \">\n\t\t<li class=\"annotations__list-item\">\n\t\t\t\t\t\t<span class=\"annotations__type d-block text-uppercase font-weight-semibold text-neutral-300 small\">Event<\/span>\n\t\t\t<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/event\/aaai-2021\/\" target=\"_self\" class=\"annotations__link font-weight-semibold text-decoration-none\" data-bi-type=\"annotated-link\" aria-label=\"Microsoft at AAAI 2021\" data-bi-aN=\"citation\" data-bi-cN=\"Microsoft at AAAI 2021\">\n\t\t\t\tMicrosoft at AAAI 2021&nbsp;<span class=\"glyph-append glyph-append-chevron-right glyph-append-xsmall\"><\/span>\n\t\t\t<\/a>\n\t\t\t\t\t<\/li>\n\t<\/ul>\n<\/div>\n\n\n\n<p>Microsoft and its researchers have been pursuing and helping to foster responsible AI for years\u2014developing innovative <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/blog\/fairness-and-interpretability-in-ai-putting-people-first\/\">AI ethics checklists and fairness assessment tools<\/a> like Fairlearn, establishing the <a href=\"https:\/\/www.microsoft.com\/en-us\/ai\/our-approach?activetab=pivot1:primaryr5\">Aether Committee<\/a> to make <a href=\"https:\/\/www.microsoft.com\/en-us\/ai\/responsible-ai?activetab=pivot1:primaryr6\">principle-based recommendations<\/a>, and laying out <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/blog\/guidelines-for-human-ai-interaction-design\/\">guidelines for human-AI interaction<\/a>, to name only a few of the milestones in this area.<\/p>\n\n\n\n<p>As a natural extension, researchers from Microsoft are presenting papers at this year\u2019s AAAI that show the wide net they\u2019re casting when it comes to developing responsible AI and using it for applications that do good. In \u201c<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/how-linguistically-fair-are-multilingual-pre-trained-language-models\/\">How Linguistically Fair are Multilingual Pre-Trained Language Models?<\/a>,\u201d researchers explore the fairness of current large multilingual language models across different languages. More specifically, they uncover how choices have been made about which models are fair and offer strategies for how these decision processes can be improved. Another paper demonstrates how AI can impact both specific industries and global challenges. In \u201c<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/where-theres-smoke-theres-fire-wildfire-risk-predictive-modeling-via-historical-climate-data\/\">Where there\u2019s Smoke, there\u2019s Fire: Wildfire Risk Predictive Modeling via Historical Climate Data<\/a>,\u201d researchers reexamine how AI can be used to predict wildfires by taking historical, climate, and geospatial data into account to improve modeling.<\/p>\n\n\n\n<p>The below&nbsp;selection&nbsp;of AAAI-accepted papers&nbsp;showcases&nbsp;specific advances with the potential&nbsp;to have&nbsp;far-reaching impact.&nbsp;AI that empowers&nbsp;all&nbsp;people&nbsp;is the end goal, whether that be&nbsp;through&nbsp;better communication, better protection of&nbsp;their&nbsp;privacy,&nbsp;or&nbsp;better optimization of&nbsp;everyday&nbsp;processes&nbsp;in specific&nbsp;fields.&nbsp;&nbsp;<\/p>\n\n\n\n<p>For more on what&nbsp;Microsoft, a&nbsp;silver sponsor of the conference, and its&nbsp;researchers are undertaking when it comes to moving AI forward, explore more at the&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/event\/aaai-2021\/#!schedule\" target=\"_blank\" rel=\"noreferrer noopener\">Microsoft at AAAI 2021<\/a>&nbsp;page.&nbsp;<\/p>\n\n\n\n<h2 id=\"policy-optimization-using-reinforcement-learning-for-universal-trading\">Policy optimization using reinforcement learning for universal trading<\/h2>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAI-Fig-1.jpg\" alt=\"disagram\" class=\"wp-image-724384\" width=\"426\" height=\"408\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAI-Fig-1-300x288.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAI-Fig-1-768x738.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAI-Fig-1-12x12.jpg 12w\" sizes=\"(max-width: 426px) 100vw, 426px\" \/><figcaption>Figure 1: In oracle policy distillation, the optimal trading strategy achieved by the teacher is used in learning the common policy (student). All the modules with dotted lines, representing teacher and policy distillation procedure, are only used during the training phase, and these would be removed during the test phase or in practice.<\/figcaption><\/figure>\n\n\n\n<p><strong>In a nutshell:<\/strong> Reinforcement learning for order execution in quantitative investment.<\/p>\n\n\n\n<p><strong>Going deeper<\/strong>: The paper \u201c<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/universal-trading-for-order-execution-with-oracle-policy-distillation\/\">Universal Trading for Order Execution with Oracle Policy Distillation<\/a>\u201d proposes a novel universal trading policy optimization framework for order execution in quantitative finance. It bridges the gap between noisy yet imperfect market states and optimal action sequences for order execution. Particularly, on one side, this framework leverages a policy distillation method that can better guide the learning of the common policy toward practically optimal execution by an oracle teacher with perfect information to approximate the optimal trading strategy. On the other side, a universal trading policy has been derived from the market data of various instruments, which is more training effective and more general to trade for different instruments.<\/p>\n\n\n\n<p><strong>Potential reach: <\/strong>This work can create an impact in the field of trading optimization in quantitative financial investment. The proposed universal learning-to-trade paradigm could substantially advance trading optimization with potentially significant profit gaining in order execution. The <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/qlib\/tree\/high-freq-execution\/examples\/trade\">code is available<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> in the Qlib project on GitHub.<\/p>\n\n\n\n<p><strong>First of its kind:<\/strong> To the best of the researchers\u2019 knowledge, this is the first work to employ policy distillation in reinforcement learning to bridge the gap between imperfect noisy data and optimal action sequences. Moreover, the work shows that direct policy optimization has a great advantage over the traditional model-based financial methods and value-based model-free reinforcement learning methods.<\/p>\n\n\n\n<p><strong>The&nbsp;people&nbsp;and organizations involved:&nbsp;<\/strong><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/kanren\/\" target=\"_blank\" rel=\"noreferrer noopener\">Kan Ren<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>,&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/weiqiliu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Weiqing Liu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>,&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/dongzho\/\" target=\"_blank\" rel=\"noreferrer noopener\">Dong Zhou<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>,&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/jibian\/\" target=\"_blank\" rel=\"noreferrer noopener\">Jiang Bian<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, and&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/tyliu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Tie-Yan Liu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>&nbsp;from Microsoft Research&nbsp;Asia;&nbsp;<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/apex.sjtu.edu.cn\/members\/arthur_fyc\" target=\"_blank\" rel=\"noreferrer noopener\">Yuchen Fang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>,&nbsp;<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"http:\/\/wnzhang.net\/\" target=\"_blank\" rel=\"noreferrer noopener\">Weinan Zhang<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>,&nbsp;and&nbsp;<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=\"noreferrer noopener\">Yong Yu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>&nbsp;from Shanghai Jiao Tong University&nbsp;<\/p>\n\n\n\n<p><strong>Additional resources and related work:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2009.11189\">Paper: \u201cQlib: An AI-oriented Quantitative Investment Platform\u201d<\/a><\/li><li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/qlib\">Qlib on GitHub<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-asia\/\">Microsoft Research Asia<\/a><\/li><\/ul>\n\n\n\n\n\t<div class=\"border-bottom border-top border-gray-300 mt-5 mb-5 msr-promo text-center text-md-left alignwide\" data-bi-aN=\"promo\" data-bi-id=\"1085508\">\n\t\t\n\n\t\t<p class=\"msr-promo__label text-gray-800 text-center text-uppercase\">\n\t\t<span class=\"px-4 bg-white display-inline-block font-weight-semibold small\">Spotlight: Blog post<\/span>\n\t<\/p>\n\t\n\t<div class=\"row pt-3 pb-4 align-items-center\">\n\t\t\t\t\t\t<div class=\"msr-promo__media col-12 col-md-5\">\n\t\t\t\t<a class=\"bg-gray-300\" href=\"https:\/\/www.microsoft.com\/en-us\/research\/blog\/medfuzz-exploring-the-robustness-of-llms-on-medical-challenge-problems\/\" aria-label=\"MedFuzz: Exploring the robustness of LLMs on medical challenge problems\" data-bi-cN=\"MedFuzz: Exploring the robustness of LLMs on medical challenge problems\" target=\"_blank\">\n\t\t\t\t\t<img decoding=\"async\" class=\"w-100 display-block\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/09\/MedFuzz-BlogHeroFeature-1400x788-1.jpg\" alt=\"MedFuzz blog hero (decorative)\" \/>\n\t\t\t\t<\/a>\n\t\t\t<\/div>\n\t\t\t\n\t\t\t<div class=\"msr-promo__content p-3 px-5 col-12 col-md\">\n\n\t\t\t\t\t\t\t\t\t<h2 class=\"h4\">MedFuzz: Exploring the robustness of LLMs on medical challenge problems<\/h2>\n\t\t\t\t\n\t\t\t\t\t\t\t\t<p class=\"large\">Medfuzz tests LLMs by breaking benchmark assumptions, exposing vulnerabilities to bolster real-world accuracy.<\/p>\n\t\t\t\t\n\t\t\t\t\t\t\t\t<div class=\"wp-block-buttons justify-content-center justify-content-md-start\">\n\t\t\t\t\t<div class=\"wp-block-button\">\n\t\t\t\t\t\t<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/blog\/medfuzz-exploring-the-robustness-of-llms-on-medical-challenge-problems\/\" class=\"btn btn-brand glyph-append glyph-append-chevron-right\" aria-label=\"Read more\" data-bi-cN=\"MedFuzz: Exploring the robustness of LLMs on medical challenge problems\" target=\"_blank\">\n\t\t\t\t\t\t\tRead more\t\t\t\t\t\t<\/a>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div><!--\/.msr-promo__content-->\n\t<\/div><!--\/.msr-promo__inner-wrap-->\n\t<\/div><!--\/.msr-promo-->\n\t\n\n\n\n<h2 id=\"uwspeech-a-speech-to-speech-translation-system-that-doesnt-rely-on-written-text\">UWSpeech: A speech-to-speech translation system that doesn\u2019t rely on written text<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"803\" height=\"426\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig2.jpg\" alt=\"diagram\" class=\"wp-image-724390\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig2.jpg 803w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig2-300x159.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig2-768x407.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig2-16x8.jpg 16w\" sizes=\"(max-width: 803px) 100vw, 803px\" \/><figcaption>Figure 2: UWSpeech process. 1) Target unwritten speech is converted into discrete tokens. 2) Source-language speech is translated into target discrete tokens. 3) Target discrete tokens are synthesized into target speech using an inverter.<\/figcaption><\/figure>\n\n\n\n<p><strong>In a nutshell: <\/strong>Want a translation system for languages with no written text? UWSpeech is your choice.<\/p>\n\n\n\n<p><strong>Going deeper: <\/strong>Existing speech-to-speech translation systems rely on the text of target language, and these existing systems can\u2019t be applied to unwritten target languages (languages without written text or phonemes). In the paper \u201c<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/uwspeech-speech-to-speech-translation-for-unwritten-languages\/\">UWSpeech: Speech to Speech Translation for Unwritten Languages<\/a>,\u201d researchers developed UWSpeech, a translation system for unwritten languages. UWSpeech converts target unwritten speech into discrete tokens with a converter. It then translates source-language speech into target discrete tokens with a translator and, finally, synthesizes target speech from target discrete tokens with an inverter. The researchers propose a method called XL-VAE in UWSpeech to enhance vector quantized variational autoencoder (VQ-VAE) with cross-lingual (XL) speech recognition, in order to train the converter and inverter of UWSpeech jointly.<\/p>\n\n\n\n<p><strong>Potential reach: <\/strong>This research sits broadly within cross-lingual speech translation, which can impact many scenarios where one spoken language needs to be translated into another. Conversations, lectures, international travel, and conferences are all examples where UWSpeech could be utilized. UWSpeech can also help to preserve unwritten languages spoken by a small amount of people.<\/p>\n\n\n\n<p><strong>Extended applications: <\/strong>Although this paper focuses on how UWSpeech can be applied to speech-to-speech translation, it can also be used to improve text-to-speech and speech-to-text translation, showing promising results in both areas. See the paper for more details.<\/p>\n\n\n\n<p><strong>The people and organizations involved:&nbsp;<\/strong><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/xuta\/\" target=\"_blank\" rel=\"noreferrer noopener\">Xu Tan<\/a>,&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/taoqin\/\" target=\"_blank\" rel=\"noreferrer noopener\">Tao&nbsp;Qin<\/a>,&nbsp;and&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/tyliu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Tie-Yan Liu<\/a>&nbsp;from the Machine Learning Group at Microsoft Research Asia;&nbsp;Chen Zhang, Yi Ren,&nbsp;and&nbsp;Kejun Zhang&nbsp;from Zhejiang University&nbsp;<\/p>\n\n\n\n<p><strong>Additional resources and related work:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/speechresearch.github.io\/\">More recent speech research in collaboration with Microsoft Research Asia<\/a> <\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/group\/deep-and-reinforcement-learning-group\/\">Deep and Reinforcement Learning Group<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/group\/machine-learning-research-group\/\">Machine Learning Group<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-asia\/\">Microsoft Research Asia<\/a><\/li><\/ul>\n\n\n\n<h2 id=\"investigating-how-data-augmentation-affects-privacy-in-deep-learning-models\">Investigating how data augmentation affects privacy in deep learning models<\/h2>\n\n\n\n<p><strong>In a nutshell: <\/strong>Watch out! Data augmentation could actually hurt privacy. Stronger membership inference attack reveals where we need to improve protection.<\/p>\n\n\n\n<p><strong>Going deeper:<\/strong> The paper \u201c<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/how-does-data-augmentation-affect-privacy-in-machine-learning\/\">How Does Data Augmentation Affect Privacy in Machine Learning?<\/a>\u201d challenges a common belief that data augmentation can prevent overfitting and hence protect the model from leakage of individual data points. The researchers developed membership inference algorithms that employ augmented instances and achieve state-of-the-art success rates of attacking well-generalized models trained with data augmentation, showing that privacy risk in these deep learning models could be greater than previously thought. Revealing this vulnerability encourages future development of techniques to strengthen the privacy protections of data augmentation as a training method.<\/p>\n\n\n\n<p><strong>Potential reach:<\/strong> The new proposed membership inference algorithms can better evaluate the privacy risk of a model and can hence help prevent other privacy attacks.<\/p>\n\n\n\n<p><strong>Toward better privacy:<\/strong> The end goal is to make a privacy guarantee in real-world machine learning tasks practical.<\/p>\n\n\n\n<p><strong>The people and organizations involved:&nbsp;<\/strong><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/huzhang\/\" target=\"_blank\" rel=\"noreferrer noopener\">Huishuai Zhang<\/a>,&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/wche\/\" target=\"_blank\" rel=\"noreferrer noopener\">Wei Chen<\/a>,&nbsp;and&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/tyliu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Tie-Yan Liu<\/a>&nbsp;of Microsoft Research Asia;&nbsp;Da Yu,&nbsp;intern at&nbsp;Microsoft Research&nbsp;Asia&nbsp;at the time of the work&nbsp;and student at&nbsp;Sun&nbsp;Yat-Sen University;&nbsp;Jian Yin, Professor at&nbsp;Sun&nbsp;Yat-Sen University&nbsp;<\/p>\n\n\n\n<p><strong>Additional resources and related work:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/do-not-let-privacy-overbill-utility-gradient-embedding-perturbation-for-private-learning\/\">Paper: &#8220;Do Not Let Privacy Overbill Utility: Gradient Embedding Perturbation For Private Learning&#8221;<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/gradient-perturbation-is-underrated-for-differentially-private-convex-optimization\/\">Paper: &#8220;Gradient Perturbation is Underrated for Differentially Private Convex Optimization&#8221;<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/group\/machine-learning-research-group\/\">Machine Learning Group<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-asia\/\">Microsoft Research Asia<\/a><\/li><\/ul>\n\n\n\n<h2 id=\"modeling-speech-and-noise-simultaneously-for-state-of-the-art-speech-enhancement\">Modeling speech and noise simultaneously for state-of-the-art speech enhancement<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"371\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-1024x371.png\" alt=\"diagram\" class=\"wp-image-724396\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-1024x371.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-300x109.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-768x278.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-1536x557.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-2048x742.png 2048w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/AAAIFig3-16x6.png 16w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><figcaption>Figure 3: SN-Net has a two-branch structure, where each branch is based on an encoder-decoder. One branch predicts speech (top), while the other predicts noise (bottom). With the addition of RA blocks in each branch, SN-Net can simultaneously mine the potential of different components of the noisy signal. Interaction modules transform and share information between the two branches, and a merge branch combines outputs and generates the final enhanced speech.<\/figcaption><\/figure>\n\n\n\n<p><strong>In a nutshell:&nbsp;<\/strong>A two-branch convolutional neural network approach to&nbsp;interactive speech and noise modeling for speech enhancement.<\/p>\n\n\n\n<p><strong>Going deeper:<\/strong> Mainstream deep learning\u2013based speech enhancement mainly predicts speech only, ignoring the characteristics of background noises. However, traditional speech enhancement methods mostly go the opposite way, that is, they model noises with an assumption on noise distributions. The result is that their generalization capability is limited. In the paper \u201c<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/interactive-speech-and-noise-modeling-for-speech-enhancement\/\">Interactive Speech and Noise Modeling for Speech Enhancement<\/a>,\u201d researchers propose the SN-Net, an interactive speech and noise modeling framework for speech enhancement, where speech and noise are simultaneously modeled in a two-branch deep neural network. Several interactions are introduced to help speech estimation benefit from noise prediction, and vice versa. As it&#8217;s challenging to model noises because of the diverse noise types, self-attention is employed in modeling both speech and noise. The SN-Net outperforms the state of the art by a large margin on several public datasets. <\/p>\n\n\n\n<p><strong>Potential reach<\/strong><strong>:<\/strong><strong>&nbsp;<\/strong>This technology can be widely&nbsp;impactful&nbsp;in applications where speech clarity is important, including video recordings, online meetings, and virtual lessons.&nbsp;The research can naturally be extended to use&nbsp;with&nbsp;the speaker separation task (see paper for more on this).&nbsp;<\/p>\n\n\n\n<p><strong>State&nbsp;of&nbsp;the&nbsp;art across multiple benchmarks:&nbsp;<\/strong>The researchers tested SN-Net against&nbsp;state-of-the-art&nbsp;models on Voice Bank + DEMAND and the Deep Noise Suppression (DNS)&nbsp;challenge&nbsp;dataset.&nbsp;Additionally, researchers&nbsp;conducted a two-speaker speech separation experiment on&nbsp;the TIMIT&nbsp;corpus, and SN-Net outperforms Conv-TasNet, the&nbsp;state-of-the-art&nbsp;method, for SDR (signal-to-distortion&nbsp;ratio) improvement and Perceptual Evaluation of Speech Quality (PESQ). See the paper for a detailed breakdown of these tests.&nbsp;&nbsp;<\/p>\n\n\n\n<p><strong>The people and organizations involved:&nbsp;<\/strong><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/xipe\/\" target=\"_blank\" rel=\"noreferrer noopener\">Xiulian Peng<\/a>&nbsp;and&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/yanlu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Yan Lu<\/a>&nbsp;from the Media Computing Group at Microsoft Research Asia; Sriram Srinivasan from Microsoft; and Chengyu Zheng and Yuan Zhang from Communication University of China&nbsp;<\/p>\n\n\n\n<p><strong>Additional resources&nbsp;and related work:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/group\/media-computing-group\/#:~:text=Media%20Computing%20Group.%20Overview.%20Overview.%20Online%20collaboration%20with,to%20deliver%20real-time,%20intelligent,%20and%20immersive%20media\">Media Computing Group<\/a><\/li><li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-asia\/\">Microsoft Research Asia<\/a><\/li><\/ul>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>The purpose&nbsp;of&nbsp;the&nbsp;Association for the Advancement of Artificial&nbsp;Intelligence, according to its bylaws, is&nbsp;twofold.&nbsp;The first is to promote research&nbsp;in the area of&nbsp;AI, and the second is to promote the&nbsp;responsible&nbsp;use&nbsp;of these types of technology.&nbsp;The&nbsp;result was a&nbsp;35th AAAI Conference on Artificial Intelligence (AAAI-21) (opens in new tab)&nbsp;schedule that&nbsp;broadens the possibilities&nbsp;of&nbsp;AI and&nbsp;is&nbsp;heavily reflective of&nbsp;a pivotal time in AI research&nbsp;when&nbsp;experts are [&hellip;]<\/p>\n","protected":false},"author":38838,"featured_media":724375,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"categories":[1],"tags":[],"research-area":[13556],"msr-region":[],"msr-event-type":[],"msr-locale":[268875],"msr-post-option":[],"msr-impact-theme":[],"msr-promo-type":[],"msr-podcast-series":[],"class_list":["post-724366","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-research-blog","msr-research-area-artificial-intelligence","msr-locale-en_us"],"msr_event_details":{"start":"","end":"","location":""},"podcast_url":"","podcast_episode":"","msr_research_lab":[199560],"msr_impact_theme":[],"related-publications":[],"related-downloads":[],"related-videos":[],"related-academic-programs":[],"related-groups":[144848,269241],"related-projects":[],"related-events":[713662],"related-researchers":[],"msr_type":"Post","featured_image_thumbnail":"<img width=\"960\" height=\"540\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-960x540.jpg\" class=\"img-object-cover\" alt=\"hexagonal AI icons on a blue background\" decoding=\"async\" loading=\"lazy\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-960x540.jpg 960w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-300x169.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-1024x576.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-768x432.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-1536x864.jpg 1536w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-2048x1152.jpg 2048w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-16x9.jpg 16w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-1066x600.jpg 1066w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-655x368.jpg 655w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-343x193.jpg 343w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-640x360.jpg 640w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-1280x720.jpg 1280w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2021\/02\/1400x788_AAAi_NoLogo_stills-1920x1080.jpg 1920w\" sizes=\"(max-width: 960px) 100vw, 960px\" \/>","byline":"","formattedDate":"February 24, 2021","formattedExcerpt":"The purpose&nbsp;of&nbsp;the&nbsp;Association for the Advancement of Artificial&nbsp;Intelligence, according to its bylaws, is&nbsp;twofold.&nbsp;The first is to promote research&nbsp;in the area of&nbsp;AI, and the second is to promote the&nbsp;responsible&nbsp;use&nbsp;of these types of technology.&nbsp;The&nbsp;result was a&nbsp;35th AAAI Conference on Artificial Intelligence (AAAI-21) (opens in new tab)&nbsp;schedule that&nbsp;broadens the&hellip;","locale":{"slug":"en_us","name":"English","native":"","english":"English"},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/724366"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/users\/38838"}],"replies":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/comments?post=724366"}],"version-history":[{"count":12,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/724366\/revisions"}],"predecessor-version":[{"id":894492,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/724366\/revisions\/894492"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/724375"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=724366"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/categories?post=724366"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/tags?post=724366"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=724366"},{"taxonomy":"msr-region","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-region?post=724366"},{"taxonomy":"msr-event-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-event-type?post=724366"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=724366"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=724366"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=724366"},{"taxonomy":"msr-promo-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-promo-type?post=724366"},{"taxonomy":"msr-podcast-series","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-podcast-series?post=724366"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}