Mediators, confounders, colliders – a crash course in causal inference
Although one would think that the basic concepts of statistics should be the same across all sciences, there is an amazing heterogeneity between fields in how statistics is taught and practiced. On…
fly51fly (@fly51fly)
논문 'Reuse your FLOPs'는 매우 오프-폴리시(prefix) 조건화 기법을 활용해 계산 자원(FLOPs)을 재활용함으로써 어려운 강화학습 문제에서 확장성을 높이는 방식을 제안합니다. FAIR(Meta) 소속 연구로, 오프-폴리시 접두사를 활용한 샘플 효율 개선과 스케일링 전략을 다룹니다.
https://x.com/fly51fly/status/2016282688597217593
#reinforcementlearning #flops #offpolicy #scaling
fly51fly (@fly51fly) on X
[LG] Reuse your FLOPs: Scaling RL on Hard Problems by Conditioning on Very Off-Policy Prefixes
A Setlur, Z Wang, A Cohen, P Rashidinejad... [FAIR at Meta] (2026)
https://t.co/OkUomOykZs
Eric Maugendre about data
sayzard