JEPA-Anything: Learning Predictive Models across Different Worlds

Researchers introduce JEPA-Anything, a domain-agnostic predictive world-modeling framework using orthogonal predictive factorization, validated across seven domains including biology and weather.

JEPA-Anything extends joint-embedding predictive architectures (JEPA) through orthogonal predictive factorization (OPF), which decomposes latent targets into complementary factors learned via dedicated pathways. The framework is evaluated across seven domains — vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather — including 10 matched dynamics tasks, forecasting of over 1,000 clinical events, and 100-step molecular rollouts across four systems. Against matched JEPA baselines it improves metrics on all 10 dynamics tasks, cuts single-intervention prediction error on Interventional Pong by 34.8%, and achieves the lowest one-step and 100-step molecular errors across the four systems. Additionally, latent orbital modes recover the Keplerian scaling exponent (slope -1.4991). A factor-nominated biological intervention received experimental support in cell co-cultures, patient-derived organoids, tumor fragments, and mice.

• Introduces orthogonal predictive factorization (OPF), extending joint-embedding predictive architectures to heterogeneous domains by decomposing latent targets into complementary factors learned via dedicated pathways
• Evaluated across seven domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather
• Improves metrics on all 10 matched dynamics tasks against matched JEPA baselines
• Cuts single-intervention prediction error on Interventional Pong by 34.8%
• Includes forecasting of over 1,000 clinical events and 100-step molecular rollouts across four systems, with the lowest one-step and 100-step molecular errors versus baselines
• Latent orbital modes recover the Keplerian scaling exponent (slope -1.4991)
• A factor-nominated biological intervention received experimental support in cell co-cultures, patient-derived organoids, tumor fragments, and mice