Abstract: Recent advances in world models for vision have been driven largely by powerful 2D generative models that predict future scene states from visual observations and planned actions, often with little or no built-in knowledge of 3D geometry or physical dynamics. In this talk, Ayush will discuss why physical inductive biases, such as explicit 3D structure, remain useful for world modeling. He will present recent work showing how incorporating geometric and physical priors into learned models leads to more generalizable behaviour, significantly more efficient inference, and better control over predictions.

Speaker Bio.: Ayush Tewari (https://ayushtewari.com/) is an Assistant Professor at the University of Cambridge. His research interests lie in visual perception, and he primarily develops methods that infer rich structured representations of the visual world from images and videos, much like the mental models humans infer to interact with and navigate their surroundings. He was previously a postdoc at MIT, and completed his PhD at the Max Planck Institute for Informatics.