AI RESEARCH
Rethinking the Comparison Unit in Sequence-Level Reinforcement Learning: An Equal-Length Paired Training Framework from Loss Correction to Sample Construction
arXiv CS.LG
•
ArXi:2604.17328v1 Announce Type: new This paper investigates the length problem in sequence-level relative reinforcement learning. We observe that, although existing methods partially alleviate length-related phenomena, a fundamental issue remains insufficiently characterized: the comparison units used during