Randall Balestriero(@randall_balestr):SIGReg for pretraining Video Foundation Models! Our LeVJEPA opens many doors... - stable recipe with a simple loss (sigreg + prediction) - no tubelet, frame aggregation, EMA, stop-gradient, .... - 20X more FLOP efficient than VJEPA1/2 pretraining

Randall Balestriero

@randall_balestr

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

Joined April 2020

242 Following 6.6K Followers

Randall Balestriero@randall_balestr

2026.08.28 12:58

SIGReg for pretraining Video Foundation Models! Our LeVJEPA opens many doors... - stable recipe with a simple loss (sigreg + prediction) - no tubelet, frame aggregation, EMA, stop-gradient, .... - 20X more FLOP efficient than VJEPA1/2 pretraining - open source + reproducible