JEPA-Anything: Learning Predictive Models across Different Worlds
作者: Taoyong Cui, Zhongyao Wang, Xinyue Xu, Weiyang Liu, Zhaochen Yu, Yuying Zhang, Qiang Gao, Mengyue Yang, Wanli Ouyang, Pheng Ann Heng, Yingcheng Wu, Zhenfei Yin, Ling Yang
分类: cs.CL
发布日期: 2026-09-17
备注: Code: https://github.com/Gen-Verse/JEPA-Anything
🔗 代码/项目: GITHUB
💡 一句话要点
提出JEPA-Anything以解决跨领域预测模型的局限性问题
🎯 匹配领域: 支柱二:RL算法与架构 (RL & Architecture)
关键词: 跨领域学习 预测模型 正交预测因子分解 智能体 动态任务
📋 核心要点
- 现有的预测模型通常局限于特定领域,缺乏跨领域的通用性,限制了智能体的适应能力。
- JEPA-Anything通过正交预测因子分解(OPF)实现了领域无关的预测模型,能够在不同系统间共享学习原则。
- 实验结果表明,JEPA-Anything在10个动态任务上均优于基线方法,并在多个领域中实现了显著的误差降低。
📝 摘要(中文)
世界建模使智能体能够预测后果、指导干预并从交互中学习。然而,现有的预测模型通常是领域特定的。本文提出JEPA-Anything,一个基于正交预测因子分解(OPF)的领域无关框架。OPF将潜在目标分解为互补因子,通过专门的路径进行学习,并在共享的预测设计中重新组合。我们在七个领域评估JEPA-Anything,包括视觉、生物学、临床轨迹、控制、分子动力学、物理场和天气。实验结果显示,JEPA-Anything在所有10个动态任务上提升了性能,并在干预预测中显著降低了误差。
🔬 方法详解
问题定义:本文旨在解决现有预测模型的领域特定性问题,现有方法在跨领域应用时表现不佳,限制了智能体的学习和适应能力。
核心思路:JEPA-Anything提出了一种基于正交预测因子分解(OPF)的框架,通过将潜在目标分解为互补因子,利用专门的学习路径进行训练,从而实现跨领域的预测能力。
技术框架:该框架包括多个模块,首先通过OPF对潜在目标进行因子分解,然后通过不同的学习路径进行训练,最后在共享的预测设计中重新组合这些因子,以实现更准确的预测。
关键创新:JEPA-Anything的核心创新在于其正交预测因子分解方法,能够有效地将不同领域的预测任务整合到一个统一的框架中,显著提升了模型的通用性和预测能力。
关键设计:在设计中,采用了特定的损失函数来优化因子学习过程,并在网络结构上实现了模块化设计,以便于不同领域的适配和扩展。具体的参数设置和网络层次结构在实验中经过调优,以确保最佳性能。
🖼️ 关键图片
📊 实验亮点
在实验中,JEPA-Anything在10个动态任务上均超越了基线JEPA方法,特别是在Interventional Pong任务中,单次干预预测误差降低了34.8%。此外,在四个系统中,其分子预测的单步和100步误差均为最低,显示出其卓越的预测能力。
🎯 应用场景
JEPA-Anything的研究成果具有广泛的应用潜力,尤其是在需要跨领域预测和干预的场景中,如医疗健康、环境监测和智能控制等领域。其通用框架能够为不同系统提供一致的预测能力,推动科学发现和智能决策的发展。
📄 摘要(原文)
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? We introduce JEPA-Anything, a domain-agnostic framework based on orthogonal predictive factorization (OPF). Extending joint-embedding predictive architectures, OPF decomposes latent targets into complementary factors, learns them through dedicated pathways, and recombines them within a shared predictive design. We evaluate JEPA-Anything across seven domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather. Experiments span representation learning, intervention prediction, out-of-distribution generalization, and long-horizon dynamics, including 10 matched dynamics tasks, forecasting of over 1,000 clinical events, and 100-step molecular rollouts across four systems. Against matched JEPA baselines, JEPA-Anything improves reported metrics on all 10 dynamics tasks and reduces single-intervention prediction error on Interventional Pong by 34.8%. It achieves the lowest one-step and 100-step molecular errors among compared methods in all four systems. Beyond prediction, a factor-nominated biological intervention receives experimental support in cell co-cultures, patient-derived organoids, tumor fragments, and mice; latent orbital modes recover the Keplerian scaling exponent with a fitted slope of -1.4991. These results support a common factorized predictive principle across heterogeneous worlds, connecting world modeling with intervention and experimentally grounded scientific discovery. Code: https://github.com/Gen-Verse/JEPA-Anything