论文官方一手
JEPA-Anything：一套配方覆盖 7 个领域的世界模型
Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields
精选理由
世界模型不用再一个领域训一个了，JEPA 拆成 4 个因子后在 7 个领域全赢 baseline，Pong 干预误差降了 34.8%。
JEPA-Anything 将 JEPA 的单一 latent 目标拆分为 4 个正交因子，每个因子配备独立 predictor。该方法在 7 个领域共 10 项 dynamics 任务上全部超越同规模的 JEPA baseline。其中在 Interventional Pong 上，干预误差降低了 34.8%。
原文 · marktechpost
Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields
JEPA-Anything splits a JEPA's single latent target into 4 orthogonal factors, each with its own predictor. Tested across 7 domains, it beat matched JEPA baselines on all 10 dynamics tasks and cut Interventional Pong intervention error by 34.8%. The post Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields appeared first on MarkTechPost .
