method · first seen 30 Oct 2026
entropy-based reward
Reward signals derived from the entropy of model token distributions to score reasoning traces.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Self-Improvement with Auxiliary Rewards: Learning from Verified Traces and Evaluating TransferPapers · 30 Oct 2026 · Minju Gwak, Alan Ritter
Latest sources
- Self-Improvement with Auxiliary Rewards: Learning from Verified Traces and Evaluating TransferPapers · 30 Oct 2026 · Minju Gwak, Alan Ritter