JEPA-Anything: Domain-Agnostic World Modeling via Orthogonal Factorization
September 16, 2026
JEPA-Anything introduces orthogonal predictive factorization (OPF) to enable a single predictive architecture to work across vision, biology, and weather domains. The framework decomposes latent targets into complementary factors to support representation learning and out-of-distribution generalization.
HOW THIS AFFECTS YOU
●
researcherThis offers a potential path toward unified world models that generalize across radically different physical and biological systems.