About me

I’m an assistant professor at the Wadhwani School of Data Science & AI and a principal investigator at the Centre for Responsible AI (CeRAI) at IIT Madras. My group studies the theory and practice of ML & AI, focusing on privacy-preserving and robust learning of LLMs and generative AI, with applications to healthcare and public good.

Previously, I was a visiting researcher (postdoc) at Google Research in the Federated Learning team. I obtained my PhD from the Paul G. Allen School of Computer Science & Engineering at the University of Washington, where I was fortunate to be advised by Zaid Harchaoui and Sham Kakade. Before that, I worked with Nina Balcan for my Master’s at Carnegie Mellon University and received an undergraduate degree from IIT Bombay.

My research has been recognized by a NeurIPS outstanding paper award and I was a 2019-20 J.P. Morgan PhD Fellow.

Contact me at [my-first-name][my-last-initial] @ dsai [dot] iitm [dot] ac [dot] in.

Research highlights

Some highlights from my previous research include:

• evaluating synthetic data generation by LLMs and generative AI [NeurIPS ‘21 Outstanding Paper Award, JMLR 2023 Best Papers Track]: Project Page
• differentially private learning, including optimization [ICLR ‘24, FOCS ‘24], auditing [NeurIPS ‘23], datasets [NeurIPS ‘23], and LLM privacy attacks [EMNLP ‘24 Oral]
• robust aggregation for federated learning [FL-ICML ‘20 Long Oral, TSP ‘22 Top 25 Downloaded Article]
• distributionally robust and differentially private federated learning [DistShift-NeurIPS ‘22 Spotlight, MLJ ‘23, AISTATS ‘23, ICLR ‘24 Spotlight]. The AISTATS paper received the ASA Student Paper Award Honorable Mention. See the project Page

News

• [November 2024] Paper on user inference attacks for LLMs accepted for an oral presentation at EMNLP 2024!
• [October 2024] Invited talk at 1st International Conference on Responsible AI for Healthcare (webpage)!
• [July 2024] Paper on efficient and near-optimal correlated noise mechanisms for streaming DP has been accepted to FOCS 2024 and an oral presentation at TPDP 2024!
• [June 2024] Gave an invited IEEE Webinar on robust federated learning (Slides Recording)!
• [May 2024] Attended ICLR to present papers on differentially private learning with correlated noise and bias-free distributionally robust learning (Spotlight).
• [Mar. 2024] Paper on robust federated learning has been identified as one of IEEE Signal Processing Society’s top 25 downloaded articles from Sept. 2022 - Sept. 2023!
• [Feb. 2024] Concluded my time as a visiting researcher (postdoc) at Google Research where I worked on various aspects of differential privacy at the user-level [final slides, Papers: ICLR ‘24, NeurIPS ‘23, NeurIPS D&B ‘23, ArXiv ‘23]. I’m grateful to the team for such a wonderful experience!
• [Dec. 2023] Paper on MAUVE Scores for Generative AI accepted to JMLR (Best Papers Track)! See the Project Page.
• [Jan. 2023] Paper on Differentially Private Superquantile Aggregation for Federated Learning accepted to the Machine Learning Journal! See the Project Page.