A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures
Basile Terver, Randall Balestriero, Megi Dervishi, David Fan, Quentin Garrido, Tushar Nagarajan, Koustuv Sinha, Wancong Zhang, Mike Rabbat, Yann LeCun, Amir Bar · 2026
Read A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures by Basile Terver, Randall Balestriero, Megi Dervishi, David Fan, Quentin Garrido, Tushar Nagarajan, Koustuv Sinha, Wancong Zhang, Mike Rabbat, Yann LeCun, Amir Bar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03604.txt.
Description
Basile Terver and 10 co-authors release EB-JEPA, an open-source library for Joint-Embedding Predictive Architectures. Their 2026 paper describes software for learning representations by predicting in feature space rather than pixel space. That energy-based approach avoids the traps of generative modeling while capturing features that downstream tasks can use.
The library's modular examples move from image self-supervised learning to video, and then to action-conditioned world models. Each is built to train on a single GPU within a few hours, aiming at teaching as much as research. Probing the image representations reaches 91 percent accuracy on CIFAR-10. The world model example plans with 97 percent success on a two-room navigation task, and ablations show which regularizers prevent collapse.
Catalog record
| CREATOR | Basile Terver, Randall Balestriero, Megi Dervishi, David Fan, Quentin Garrido, Tushar Nagarajan, Koustuv Sinha, Wancong Zhang, Mike Rabbat, Yann LeCun, Amir Bar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03604