Vai al contenuto
AI.info

events

Cosmin goes over Score Centering Stabilizes Off-policy Reinforcement Learning

Una sessione online di un gruppo di studio in cui Cosmin esamina un articolo sulla centratura dei punteggi per stabilizzare l’apprendimento per rinforzo off-policy.

Cosmin goes over Score Centering Stabilizes Off-policy Reinforcement Learning | Eventi sull’IA

In inglese

Starts
2026-10-28 01:30 UTC
Organizer
Deep Learning Study Group (San Francisco)
Location
Online event

The Deep Learning Study Group in San Francisco holds an online session in which Cosmin walks through the paper "Score Centering Stabilizes Off-policy Reinforcement Learning," a method used in a recent update to the Thinking Machines RL platform. The session connects the paper to the Thinker StableReinforce objective and points to further reading on KL approximation and related work. It is a paper-reading discussion for people who follow reinforcement learning research and how it is implemented in training platforms.

Event website