Colloquium Series: Sinho Chewi

Our upcoming event for the Statistics Colloquium Series is scheduled for Monday, March 9 from 12:00 – 1:00pm (ET) and will be an in-person presentation at Maxwell-Dworkin 134A/B. Lunch will be provided to guests following the talk. This week's speaker will be Sinho Chewi, Assistant Professor of Statistics and Data Science at Yale.

Blind denoising diffusion models and the blessings of dimensionality

We provide a mathematical theory for diffusion models with blind denoisers—that is, score function estimators which do not take as an input the noise level along the diffusion process. We show that the noise level can be accurately estimated from a single noisy sample, provided that the intrinsic dimension of the data is substantially smaller than the ambient dimension. Consequently, we show that blind denoising diffusion models implicitly track a certain noise schedule along the diffusion, allowing us to justify their correctness. This is based on joint work with Zahra Kadkhodaie, Aram-Alexandre Pooladian, and Eero Simoncelli.

 Sinho Chewi is an Assistant Professor in the Statistics & Data Science Department at Yale University. He works on log-concave sampling, statistical applications of optimal transport, and theory for diffusion generative models. He obtained his PhD from the Massachusetts Institute of Technology.