About me

Welcome to my page! I’m a senior research scientist at Google in New York. My research is broadly on deep learning and statistical learning. I currently lead an effort on efficiently predicting data value in foundation models (with applications to Gemini and Gemma) and research fundamental techniques for model compression. I’ve also been managing Google-sponsored research collaborations with universities for the past three years.

My research has been recognized with best paper awards and honorable mentions from INFORMS ICS (2023), KDD (2022), INFORMS IOS (2020), INFORMS ICS (2020), MIT (2020), MIP Workshop (2019).

I completed my PhD at MIT where I was advised by Rahul Mazumder and worked on scalable algorithms for sparse learning. Before that, I did my masters at UIUC where I worked with ChengXiang Zhai on improving information recall in search engines.

Current research

• Data value in foundation models
  • Designing efficient methods for predicting the impact of any training data mixture on the performance of a given model (without training)
  • Researching principled statistical methods to improve the design of data ablations
  • Our approaches are being used in the development of Gemini and Gemma
• Model compression and efficiency
  • Pruning transformer-based foundation models (focus on reducing latency)
  • Improving routing in mixture of experts
  • Feature selection (best subset selection) in standard statistical learning