Research

My interests primarily lie in statistical learning theory and machine learning theory.

I work on statistical learning theory with Nikita Zhivotovskiy in Berkeley's Statistical Machine Learning group, and on deep learning theory with Michael DeWeese at BAIR.

I also work on computational nuclear physics at LLNL.

Google Scholar: link. ORCID: 0009-0004-1252-1679.

Papers

Majority-of-Three is Optimal

Divit Rawal, Nikita Zhivotovskiy

Preprint. [arXiv]

We prove that a majority vote of three independent ERMs is an optimal PAC learner, resolving the conjecture of [AHL+24]. (tldr)

A Theory of Saddle Escape in Deep Nonlinear Networks

Divit Rawal, Michael R. DeWeese

Preprint. Under review. [arXiv]

We derive the analog conservation law of balanced initialization in deep linear networks for nonlinear networks and use this to predict training dynamics for small-init networks. (tldr)

Rao-Blackwellized Score Matching on Manifolds

Divit Rawal

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

We show that denoising score matching when the latent law is supported on a manifold recovers the intrinsic Riemannian score up to a computable geometric correction. (tldr)

Minimax Rates for Hyperbolic Hierarchical Learning

Divit Rawal, Sriram Vishwanath

Preprint. [arXiv]

We prove that hierarchical data is optimally embedded in hyperbolic space.

ALPHANSO: Open-Source Modeling of (\(\alpha\),n) Neutron Source Terms

Divit Rawal, Anthony J. Nelson, William Zywiec, Daniel Siefman

Nuclear Instruments and Methods in Physics Research Section A. [doi] / [GitHub]

American Nuclear Society Student Conference, 2026. [poster]

└ Best Paper: Mathematics, Computation, and AI Applications.

Institute of Nuclear Materials Management Annual Meeting, 2026.

Invited talks at LBNL, LLNL, UT Knoxville.

A code for computing (\(\alpha\),n) neutron source terms.

I have also contributed to the development and releases of Foundation-Sec-8B-Instruct and Foundation-Sec-8B during an internship at Cisco Foundation AI, and worked on off-shell Higgs production via neural simulation-based inference with ATLAS and neural simulation-based inference for parameter estimation in ATLAS as a researcher in the Whiteson lab.