Saved in:
Bibliographic Details
Main Authors: Qu, Helen, Morel, Rudy, McCabe, Michael, Bietti, Alberto, Lanusse, François, Ho, Shirley, LeCun, Yann
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.13227
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915860084424704
author Qu, Helen
Morel, Rudy
McCabe, Michael
Bietti, Alberto
Lanusse, François
Ho, Shirley
LeCun, Yann
author_facet Qu, Helen
Morel, Rudy
McCabe, Michael
Bietti, Alberto
Lanusse, François
Ho, Shirley
LeCun, Yann
contents Machine learning approaches to spatiotemporal physical systems have primarily focused on next-frame prediction, with the goal of learning an accurate emulator for the system's evolution in time. However, these emulators are computationally expensive to train and are subject to performance pitfalls, such as compounding errors during autoregressive rollout. In this work, we take a different perspective and look at scientific tasks further downstream of predicting the next frame, such as estimation of a system's governing physical parameters. Accuracy on these tasks offers a uniquely quantifiable glimpse into the physical relevance of the representations of these models. We evaluate the effectiveness of general-purpose self-supervised methods in learning physics-grounded representations that are useful for downstream scientific tasks. Surprisingly, we find that not all methods designed for physical modeling outperform generic self-supervised learning methods on these tasks, and methods that learn in the latent space (e.g., joint embedding predictive architectures, or JEPAs) outperform those optimizing pixel-level prediction objectives. Code is available at https://github.com/helenqu/physical-representation-learning.
format Preprint
id arxiv_https___arxiv_org_abs_2603_13227
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Representation Learning for Spatiotemporal Physical Systems
Qu, Helen
Morel, Rudy
McCabe, Michael
Bietti, Alberto
Lanusse, François
Ho, Shirley
LeCun, Yann
Machine Learning
Computer Vision and Pattern Recognition
Machine learning approaches to spatiotemporal physical systems have primarily focused on next-frame prediction, with the goal of learning an accurate emulator for the system's evolution in time. However, these emulators are computationally expensive to train and are subject to performance pitfalls, such as compounding errors during autoregressive rollout. In this work, we take a different perspective and look at scientific tasks further downstream of predicting the next frame, such as estimation of a system's governing physical parameters. Accuracy on these tasks offers a uniquely quantifiable glimpse into the physical relevance of the representations of these models. We evaluate the effectiveness of general-purpose self-supervised methods in learning physics-grounded representations that are useful for downstream scientific tasks. Surprisingly, we find that not all methods designed for physical modeling outperform generic self-supervised learning methods on these tasks, and methods that learn in the latent space (e.g., joint embedding predictive architectures, or JEPAs) outperform those optimizing pixel-level prediction objectives. Code is available at https://github.com/helenqu/physical-representation-learning.
title Representation Learning for Spatiotemporal Physical Systems
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.13227