AI RESEARCH
Reference-Sampled Boltzmann Projection for KL-Regularized RLVR: Target-Matched Weighted SFT, Finite One-Shot Gaps, and Policy Mirror Descent
arXiv CS.LG
•
ArXi:2605.02469v1 Announce Type: new Online reinforcement learning with verifiable rewards (RLVR) turns checkable outcomes into a scalable