{"ID":2843565,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.08544","arxiv_id":"2511.08544","title":"LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics","abstract":"Learning manipulable representations of the world and its dynamics is central to AI. Joint-Embedding Predictive Architectures (JEPAs) offer a promising blueprint, but lack of practical guidance and theory has led to ad-hoc R\u0026D. We present a comprehensive theory of JEPAs and instantiate it in {\\bf LeJEPA}, a lean, scalable, and theoretically grounded training objective. First, we identify the isotropic Gaussian as the optimal distribution that JEPAs' embeddings should follow to minimize downstream prediction risk. Second, we introduce a novel objective--{\\bf Sketched Isotropic Gaussian Regularization} (SIGReg)--to constrain embeddings to reach that ideal distribution. Combining the JEPA predictive loss with SIGReg yields LeJEPA with numerous theoretical and practical benefits: (i) single trade-off hyperparameter, (ii) linear time and memory complexity, (iii) stability across hyper-parameters, architectures (ResNets, ViTs, ConvNets) and domains, (iv) heuristics-free, e.g., no stop-gradient, no teacher-student, no hyper-parameter schedulers, and (v) distributed training-friendly implementation requiring only $\\approx$50 lines of code. Our empirical validation covers 10+ datasets, 60+ architectures, all with varying scales and domains. As an example, using imagenet-1k for pretraining and linear evaluation with frozen backbone, LeJEPA reaches 79\\% with a ViT-H/14. We hope that the simplicity and theory-friendly ecosystem offered by LeJEPA will reestablish self-supervised pre-training as a core pillar of AI research (\\href{https://github.com/rbalestr-lab/lejepa}{GitHub repo}).","short_abstract":"Learning manipulable representations of the world and its dynamics is central to AI. Joint-Embedding Predictive Architectures (JEPAs) offer a promising blueprint, but lack of practical guidance and theory has led to ad-hoc R\u0026D. We present a comprehensive theory of JEPAs and instantiate it in {\\bf LeJEPA}, a lean, scala...","url_abs":"https://arxiv.org/abs/2511.08544","url_pdf":"https://arxiv.org/pdf/2511.08544v3","authors":"[\"Randall Balestriero\",\"Yann LeCun\"]","published":"2025-11-11T18:21:55Z","proceeding":"cs.LG","tasks":"[\"cs.LG\",\"cs.AI\",\"cs.CV\",\"stat.ML\"]","methods":"[]","has_code":false,"code_links":[{"ID":607227,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2843565,"paper_url":"https://arxiv.org/abs/2511.08544","paper_title":"LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics","repo_url":"https://github.com/rbalestr-lab/lejepa","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}