Randall Balestriero's banner
Randall Balestriero's profile picture

Randall Balestriero

@randall_balestr • 7,136 subscribers

AI Researcher: From theory to practice (and back) Postdoc @MetaAI with @ylecun PhD @RiceUniversity with @rbaraniuk Masters @ENS_Ulm @Paris_Sorbonne

Videos

LIVE
1.2k

Anya Rossi

sweetdream.ai

SweetDream.aiSponsored•Livecam
Streaming Now

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

HD live stream
Exclusive private shows
1.2k viewers online

Free preview available • Premium content

The Platonic Representation Hypothesis is now a Theorem ◼️ Train JEPAs separately, on any modalities: they provably learn the same representation up to one rotation. The result? A 109M JEPA-Text model that aligns with vision models better than 72B LLMs. w/ Yann LeCun 🧵
1:45
randall_balestr's profile picture

The Platonic Representation Hypothesis is now a Theorem ◼️ Train JEPAs separately, on any modalities: they provably learn the same representation up to one rotation. The result? A 109M JEPA-Text model that aligns with vision models better than 72B LLMs. w/ Yann LeCun 🧵

Randall Balestriero

49,838 просмотров • 16 часов назад

LeJEPA: a novel pretraining paradigm free of the (many) heuristics we relied on (stop-grad, teacher, ...) - 60+ arch., up to 2B params - 10+ datasets - in-domain training (>DINOv3) - corr(train loss, test perf)=95% Paper: Code:
0:28
randall_balestr's profile picture

LeJEPA: a novel pretraining paradigm free of the (many) heuristics we relied on (stop-grad, teacher, ...) - 60+ arch., up to 2B params - 10+ datasets - in-domain training (>DINOv3) - corr(train loss, test perf)=95% Paper: Code:

Randall Balestriero

506,032 просмотров • 11 месяцев назад

World Modeling research needs fast iteration, reproducibility, optimized baselines, open-source, and precise zero-shot stress testing. Here comes stable-worldmodel! Paper: Code: Come stress-test your model/idea! DINO-WM results ⬇️
0:43
randall_balestr's profile picture

World Modeling research needs fast iteration, reproducibility, optimized baselines, open-source, and precise zero-shot stress testing. Here comes stable-worldmodel! Paper: Code: Come stress-test your model/idea! DINO-WM results ⬇️

Randall Balestriero

44,300 просмотров • 8 месяцев назад

Больше нет контента для загрузки