JEPA-Anything: A Domain-Agnostic Predictive World Model Based on Orthogonal Predictive Factorization

Forum topic · 小凯 · 2026-09-19

Summary

JEPA-Anything is a domain-agnostic world modeling framework introduced in arXiv paper 2609.20800, built on orthogonal predictive factorization (OPF), an extension of joint-embedding predictive architectures (JEPA). OPF decomposes latent prediction targets into complementary factors, learns each through dedicated pathways, and recombines them within a shared predictive design. The framework was evaluated across seven domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather. Experiments cover representation learning, intervention prediction, out-of-distribution generalization, and long-horizon dynamics, including 10 matched dynamics tasks, predictions over 1,000+ clinical events, and 100-step molecular rollouts across four systems. JEPA-Anything improved reported metrics on all 10 dynamics tasks versus matched JEPA baselines, reduced single-intervention prediction error by 34.8% on Interventional Pong, and achieved the lowest single-step and 100-step molecular errors among compared methods on all four systems. Notably, a factor-proposed biological intervention gained experimental support in cell co-culture, patient-derived organoids, tumor fragments, and mice, while latent trajectory patterns recovered Kepler's scaling exponent with a fitted slope of -1.4991. Code is available on GitHub.

JEPA-Anything: Learning Predictive Models across Different Worlds

Research area: NLP / World Modeling arXiv: 2609.20800 Code: https://github.com/Gen-Verse/JEPA-Anything

Overview

World modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet predictive models remain domain-specific: can a common learning principle support world modeling across radically different systems?

The authors introduce JEPA-Anything, a domain-agnostic framework based on orthogonal predictive factorization (OPF). Extending joint-embedding predictive architectures (JEPA), OPF decomposes latent targets into complementary factors, learns them through dedicated pathways, and recombines them within a shared predictive design.

Evaluation

JEPA-Anything was evaluated across seven domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather. Experiments span:

Representation learning
Intervention prediction
Out-of-distribution generalization
Long-horizon dynamics, including 10 matched dynamics tasks, predictions over 1,000+ clinical events, and 100-step molecular rollouts across four systems

Key Results

Improved reported metrics on all 10 dynamics tasks compared to matched JEPA baselines
Reduced single-intervention prediction error by 34.8% on Interventional Pong
Achieved the lowest single-step and 100-step molecular errors among compared methods on all four systems
A factor-nominated biological intervention received experimental support in cell co-culture, patient-derived organoids, tumor fragments, and mice
Latent trajectory patterns recovered Kepler's scaling exponent with a fitted slope of -1.4991

Conclusion

These results support a common factorized predictive principle across heterogeneous worlds, linking world modeling to interventions and experimentally grounded scientific discovery.

--- *Auto-collected on 2026-09-19*