I am currently a Senior Research Scientist at Google Research.
My research explores the foundations of large-scale machine learning models. Recently, I have been working on improving text-to-image generation models with VLM-based feedback. I'm also interested in robustness and stability, including adversarial attacks and OOD generalization. Finally, I enjoy formulating new theoretical ML models, such as for sparsely activated mixture-of-experts networks, and new probing tasks for the equivariance of image embeddings.
Prior to Google, I was a postdoc at UCSD in the CSE department, working with Kamalika Chaudhuri, Sanjoy Dasgupta, and Paul Siegel. Before this, I was a visiting Research Scientist at Facebook Reality Labs, exploring Distributed Computing and Augmented Reality. In June 2018, I received my PhD in Computer Science & Engineering from the University of Washington (UW). My advisor was Paul Beame, and I was a part of both the Theory and MISL groups. I also spent time at Microsoft Research, working with the DNA Storage and the Theory groups. I obtained my BS from the Department of Computer Science at the University of Illinois Urbana-Champaign (UIUC).
For more information, take a look at my Publications. See also the UCSD Machine Learning Blog for expository articles about some previous papers and more.
News
(Dec 2023) Paper at NeurIPS 2023 on Benchmarking Robustness to Adversarial Image Obfuscations. Check out our dataset for a challenging new obfuscation-based robustness challenge!
(Dec 2023) Paper at DistShift'23 @ NeurIPS 2023 on probing image embeddings with transformation prediction. We find compare several SSL embedding in terms of their non-semantic visual content.
(Nov 2023) Check out DreamSync our new approach to improve text-to-image generation models for both prompt alignment and visual aesthetics!