World Models: Discovering and Simulating Physics at Scale

Organizers:

Randall Balestriero, Brown University
Shirley Ho, Flatiron Institute
Julia Kempe, New York University
Yann LeCun, New York University 

Physics has always advanced by distilling complex phenomena into compact, predictive descriptions — effective theories, conservation laws, and symmetries. Yet modern experiments and simulations now produce vast streams of high-dimensional data — time-resolved images, particle tracks, turbulent flows, astronomical surveys, and molecular trajectories — that can no longer be understood by hand In parallel, advances in self-supervised learning, world models, and Joint Embedding Predictive Architectures (JEPAs) have created powerful general-purpose tools for learning predictive, invariant structure directly from data, at scales and in regimes where classical analytical approaches struggle. Bringing together researchers in theoretical and experimental physics, machine learning, and the quantitative sciences, the program centers on three themes — self-supervised learning as a principle of scientific inference; world models as a unifying framwork for physical systems; and the automated discovery of physics structure from high-dimensional data. Sessions will mix invited and contributed talks, poster sessions with strong early-career participation, and afternoon breakout discussions, with a sustained emphasis on open science—shared code, public datasets, and common benchmarks.

Learn more at the conference website: https://wmw-aspen.github.io/