Budhaditya Deb
Principal Scientist
About
I am a Principal Scientist in the Microsoft Search, Assistant and Intelligence (MSAI) group. I completed my Ph.D from Rutgers University in 2005. Prior to Microsoft, I spent 8 years at GE-Research and BBN Technologies as a Researcher working on various industrial, academic and government projects. My current research interests span areas under Natural Language Processing and Understanding, Bayesian DNNs, Semi-Supervised Learning and Transfer Learning.
I have been the core contributor behind several AI features for MS-Office in production. These include:
- Classification and Information Extraction Framework for Text (CIFT), a system for generating actionable insights from large-scale unstructured text signals.
- Domain Classifiers, Commitments, and Requests models for Outlook emails.
- SmartReply for Outlook, Skype and Teams: I have authored the core DNN models for all versions of SR shipped in production. I also lead the research and development behind SmartReply internationalization efforts.
- Meeting catchup and summarization in Microsoft Teams, which provides insights, notes and summary from Teams meetings transcripts.
My current research project involves developing interactive Large Language Models which have intrinsic ability to continuously interact with users and learn in natural language.
Featured content
Assistive AI Makes Replying Easier
Microsoft’s mission is to empower every person and organization to achieve more. So, we are constantly looking for opportunities to simplify workflows and save people time and effort. Sending replies to email or chat messages is a common activity and people spend considerable amount of time on it. By harnessing the power of AI, we are helping people reply faster by intelligently suggesting replies which can be used to easily respond to messages with a simple click or tap on the device. For email messages, people can then edit the response or hit the ‘Send’ button, while for chat messages, the reply is immediately sent. The difference in behavior is designed keeping in mind that for chats, people can break their reply content into a few adjacent chat snippets and so immediately sending the reply offers the quickest workflow. We have also expanded the feature to multiple international languages like Spanish, Portuguese, French and Italian and plan to roll out to several new languages and markets in the next year. This feature currently saves users hundreds of millions of keystrokes each month.
Diversifying Reply Suggestions Using a Matching-Conditional Variational Autoencoder
Delivering high quality features to our users often requires developing new techniques. We also believe that participating in academic research helps us to be more rigorous while contributing to the larger community. This paper from NAACL 2019 describes a technique that increases the diversity of reply suggestions, helping more users to find a suggestion that they can use. Great work Budhaditya Deb, Peter Bailey and Milad Shokouhi!
Microsoft Teams Premium arrives with meeting recaps and live translations
Microsoft is launching Microsoft Teams Premium today, a new offering designed to boost its chat and collaboration app with some AI-powered smart features. The biggest addition is a new meeting recap feature that essentially adds an AI virtual assistant to each meeting to share highlights and automatically capture all of the important parts. It’s like having your own assistant to take notes for you, even if you’re not there.