Sabarish Sainathan

About

Sabarish Sainathan studies the scaling and generalization behavior of foundation models in order to develop competitive models that are smaller, more efficient, or more aligned with theory. He is a PHD Student in Computer Science at Harvard John A. Paulson School of Engineering and Applied Sciences. He is interested in studying large learning models through the lens of probability. Previously, he graduated with a BSE in computer science from Princeton, where he studied the properties of kernels associated with neural networks.

Research Focus

Sainathan’s current research focuses on neural network dynamics and continuous-time stochastic processes. He studies the properties of mean-field, infinite-width limits of neural networks to better answer how to scale neural networks. He is also broadly interested in theoretical properties of diffusion processes: their convergence properties; how score estimation contributes to generalization; and how they may be distilled into more compact learned representations.