JEPA-Anything: Learning Predictive Models across Different Worlds
JEPA-Anything reports one predictive framework working across seven very different domains, from weather to clinical trajectories.
The paper introduces orthogonal predictive factorization, which splits latent targets into complementary factors and recombines them inside a shared JEPA-style model. Its tests cover representation learning, interventions, out-of-distribution cases, and long-horizon dynamics. Against matched JEPA baselines, it improves reported metrics on all 10 dynamics tasks and cuts Interventional Pong single-intervention error by 34.8%. The authors also report experimental support for a factor-nominated biological intervention across several biological systems. HF Daily Papers' note
The paper introduces orthogonal predictive factorization, which splits latent targets into complementary factors and recombines them inside a shared JEPA-style model. Its tests cover representation learning, interventions, out-of-distribution cases, and long-horizon dynamics. Against matched JEPA baselines, it improves reported metrics on all 10 dynamics tasks and cuts Interventional Pong single-intervention error by 34.8%. The authors also report experimental support for a factor-nominated biological intervention across several biological systems. HF Daily Papers' note
score 6