Mozes Jacobs

About

Mozes Jacobs earned his BS in Computer Science at the University of Washington (UW), Seattle in 2022. At UW, he completed his undergraduate honors thesis with Professor Rajesh Rao. Inspired by the predictive coding theory of cortical function, Jacobs developed a variational autoencoder method for videos that performed inference using prediction error signals. In addition, he worked under Professor William Noble on PASTIS, a Python package that infers 3D chromatin structures from matrices of experimental data. After graduating, he worked with Professor Nathan Kutz and Dr. Ryan Raut as a Shanahan Foundation Postbac Fellow. He developed HyperSINDy, a framework for modeling stochastic dynamics via a generative model of sparse governing equations.

Research Focus

Jacobs’ research currently focuses on computer vision and generative modeling. His interests are broadly in computer vision and generative modeling. He is particularly intrigued by problems like image generation (e.g. text-to-image diffusion models), video generation, and scene understanding. He is interested in building systems that can fully understand visual scenes, which involves being able to decompose scenes into objects and their relationships. Jacobs is curious about how we can bake inductive biases into generative models for these problems.