harvard art museums

Divit Rawal

I'm an undergrad at Berkeley studying statistics, computer science, and physics.

I'm generally interested in statistically efficient learning in structured settings. I work on statistical learning theory in Berkeley's StatML group with Nikita Zhivotovskiy, and deep learning theory at BAIR with Michael DeWeese.

Selected Publications (full list at research)

Majority-of-Three is Optimal [arXiv]

with N. Zhivotovskiy. Under review.

A Theory of Saddle Escape in Deep Nonlinear Networks [arXiv]

with M. R. DeWeese. Under review.

Rao-Blackwellized Score Matching on Manifolds [arXiv]

ICML 2026, Workshop on Structured Probabilistic Inference and Generative Modeling.