AI RESEARCH

A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures

arXiv CS.CV

ArXi:2602.03604v3 Announce Type: replace We present EB-JEPA, an open-source library for learning representations and world models using Joint-Embedding Predictive Architectures (JEPAs). JEPAs learn to predict in representation space rather than pixel space, avoiding the pitfalls of generative modeling while capturing semantically meaningful features suitable for downstream tasks.