works on
advantage shaping 1contrastive learning 1on-policy distillation 1policy optimization 1token-level correctness 1

From the 1 of 4 linked papers with an AI index.

collaborators
Showing cs.LGShow all

1 paper · 1 filter

cs.LG2026

Beyond Entropy: Correctness-Aware Advantage Shaping via Contrastive Policy Optimization

Weiwen Xu, Jia Liu, Hou Pong Chan +4

The paper proposes Contrastive Policy Optimization, which leverages token‑level contrastive disagreement between reference‑guided and standard generation distributions to provide a…