top | item 45903672

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

5 points| XTXinverseXTY | 3 months ago |arxiv.org

2 comments

order

rvz|3 months ago

A breath of fresh air from the over-concentration on LLM research. Something different for once.

XTXinverseXTY|3 months ago

Seems like this signals Yann Lecun's direction now that he's leaving Meta

The EMA teacher model still seems like black magic to me (present in DINO and JEPA series but gone now)