{"version":1,"type":"story","url":"https://digestai.news/story/phai-labs-introduces-jepa-anything-framework-for-cross-domain-world-mo","json":"https://digestai.news/story/phai-labs-introduces-jepa-anything-framework-for-cross-domain-world-mo.json","markdown":"https://digestai.news/story/phai-labs-introduces-jepa-anything-framework-for-cross-domain-world-mo.md","slug":"phai-labs-introduces-jepa-anything-framework-for-cross-domain-world-mo","headline":"PhAI Labs introduces JEPA-Anything framework for cross-domain world models","summary":"Researchers from PhAI Labs, CUHK, Fudan, Stanford, Oxford, and Princeton released **JEPA-Anything**, a domain-agnostic framework that applies a single learning recipe across seven fields: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather. The method extends joint-embedding predictive architectures (JEPAs) with **Orthogonal Predictive Factorization (OPF)**, splitting a monolithic target embedding into orthogonal factors to avoid conflicting gradients and improve stability.\n\nJEPA-Anything outperformed matched JEPA baselines on 10 dynamics tasks, including a **34.83% reduction** in single-intervention error on Interventional Pong. In scientific analysis, it identified **IL-18 plus CD73 blockade** as a potential cancer intervention, validated in wet-lab tests. The core code is licensed under **Apache-2.0**, with research checkpoints available on Hugging Face. Results vary by task—planning gains favor Walker2d and HalfCheetah but lag in Hopper—while rollout errors dropped **~44.7%** on APEBench Burgers.","keyPoints":["JEPA-Anything applies one shared learning recipe across seven domains, improving stability via orthogonal predictive factorization","Outperformed matched JEPA baselines on 10 dynamics tasks, including a 34.83% error reduction in Interventional Pong","Core code is Apache-2.0 licensed; research checkpoints available on Hugging Face, with wet-lab validation for cancer intervention"],"whyItMatters":"The framework could reduce the need for bespoke models per domain, accelerating cross-disciplinary AI applications in simulation, forecasting, and scientific discovery.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":["PhAI Labs","Marktechpost AI Media Inc."],"models":["JEPA-Anything","I-JEPA","V-JEPA 2","Cell-JEPA","TrajCast"],"people":["Asif Razzaq"]},"firstPublishedAt":"2026-10-06T05:12:01Z","updatedAt":"2026-10-06T05:12:01Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"MarkTechPost","title":"Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields","url":"https://marktechpost.com/2026/10/05/beyond-domain-specific-world-models-jepa-anything-uses-1-recipe-for-7-fields","publishedAt":"2026-10-06T05:12:01Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"PhAI Labs introduces JEPA-Anything framework for cross-domain world models\", 6 October 2026, https://digestai.news/story/phai-labs-introduces-jepa-anything-framework-for-cross-domain-world-mo","publisher":"Digest AI","title":"PhAI Labs introduces JEPA-Anything framework for cross-domain world models","datePublished":"2026-10-06T05:12:01Z","url":"https://digestai.news/story/phai-labs-introduces-jepa-anything-framework-for-cross-domain-world-mo"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}