method · first seen 14 Oct 2026
majority-vote reward
Also: self-labeling, self-consistency reward
Using the model's most common answer across its own samples as a pseudo-label reward for reinforcement learning without gold labels.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Wrong Votes, Right Gradients: Why Majority-Vote Rewards Track Verifiable RLPapers · 14 Oct 2026 · Dylan P. Wilson, Chengsong Huang, Jiaxin Huang
- Collective Dynamics of Self-Reinforcement: How Cross-Task Coupling Drives Policy Collapse in Unsupervised RLVRPapers · 27 Oct 2026 · Peize Ding
- The Race Is Decided Early! A Kernel Law Predicts Lock-In in Majority-Vote Test-Time RLPapers · 8 Nov 2026 · Jaeho Hwang, Sehong Park, Daejin Choi, Uiwon Hwang
Latest sources
- The Race Is Decided Early! A Kernel Law Predicts Lock-In in Majority-Vote Test-Time RLPapers · 8 Nov 2026 · Jaeho Hwang, Sehong Park, Daejin Choi, Uiwon Hwang
- Collective Dynamics of Self-Reinforcement: How Cross-Task Coupling Drives Policy Collapse in Unsupervised RLVRPapers · 27 Oct 2026 · Peize Ding
- Wrong Votes, Right Gradients: Why Majority-Vote Rewards Track Verifiable RLPapers · 14 Oct 2026 · Dylan P. Wilson, Chengsong Huang, Jiaxin Huang