Hossein Resani, Javen Qinfeng Shi · 2026 · arXiv
DOI: 10.48550/arXiv.2610.31194 · open-access PDF
Joint-embedding predictive architectures (JEPAs) mempelajari model dunia di latent space, memungkinkan perencanaan yang efisien tetapi meninggalkan keadaan mereka tidak jelas. Kami memperkenalkan Inside-JEPA, yang menempatkan saluran relasional untuk pasangan entitas teratur di dalam latent state JEPA dan menunjukkan bahwa hilangnya reverse-dynamics dapat mendongkrak saluran ini tanpa label relasional ketika kepala membaca mereka, baik sendiri atau sebagai bagian dari keadaan penuh.
Machine translation (Meta M2M100); the English original follows.
Joint-embedding predictive architectures (JEPAs) learn world models in latent space, enabling efficient planning but leaving their states opaque. We introduce Inside-JEPA, which places relational channels for ordered entity pairs inside the JEPA latent state and shows that an inverse-dynamics loss can ground these channels without relation labels when its head reads them, either alone or as part of the full state. On Something-Something V2, latent prediction alone performs no better than a random relation head, while inverse-dynamics supervision that recovers actions from channel changes impro
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.31194, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange