按年被提到
20252025 · 3 次2026
机器学习与深度学习›优化与可微编程·arXiv:2511.08544 · cs.LG

LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

Randall Balestriero, Yann LeCun · 2025
开始阅读 ↗PDFalphaXiv 讨论
被提及 · 4 人首次 2025-11-12 · 最近 2026-07-10
4不同的人
2活跃月份
4讨论串
30被引(OpenAlex)
arXiv 摘要(候选层不写导读)

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&D. 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}).

候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。

HN 书评

推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及
评论feelingsonice · 2026-07

“LeCun's new paper takes a new approach to building JEPA, which he named LeJEPA -- https://arxiv.org/abs/2511.08544 It's not that different in the underlying theory but he removes the EMA and the twin tower, makes the whole thing more straightforward.”

The Annotated JEPA原评论 ↗
被提交为 story · 3 次