Detecting and Mitigating the Correct-Answer Extinction Window in Test-Time Reinforcement Learning with Majority Voting

• URL: http://arxiv.org/abs/2605.19444v2
• Date: Wed, 27 May 2026 03:12:51 GMT
• Title: Detecting and Mitigating the Correct-Answer Extinction Window in Test-Time Reinforcement Learning with Majority Voting
• Abstract summary: Test-time reinforcement learning (TTRL) reports substantial accuracy gains on mathematical reasoning benchmarks using majority vote as a pseudo-label signal.<n>We propose TTRL-Guard, a lightweight framework with three mechanisms targeting the textitCorrect-Answer Extinction Window<n>Experiments across three models and four benchmarks show that TTRL-Guard achieves the best average pass@1 on Qwen2.5-7B-Instruct and Qwen3-4B, improves relatively over TTRL by +54% on AIME 2025.
• Score: 4.336665585098371
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Test-time reinforcement learning (TTRL) reports substantial accuracy gains on mathematical reasoning benchmarks using majority vote as a pseudo-label signal. We argue these gains are systematically misinterpreted: most reflect sharpening of already-solvable problems rather than genuine learning, while problems corrupted from correct to incorrect outnumber truly learned ones, and this damage is irreversible once majority vote locks onto a wrong answer. Per-problem tracking reveals that correct-answer signals in low-ability problems are briefly active before being permanently suppressed, a phenomenon we term the \textit{Correct-Answer Extinction Window}, with Flip Rate (FR) as its leading indicator. We thus propose TTRL-Guard, a lightweight framework with three mechanisms targeting the extinction window: Flip-Rate-Aware Reward Scaling (FRS) down-weights at-risk updates as FR declines, Minority-Preserving Sampling (MPS) retains gradient signal from minority correct answers, and Risk-Conditioned Sparse Updatings (RCSU) suspends updates on polarized problems. Experiments across three models and four benchmarks show that TTRL-Guard achieves the best average pass@1 on Qwen2.5-7B-Instruct and Qwen3-4B, improves relatively over TTRL by +54\% on AIME 2025.

Related papers

• TTPO: Test-Time Policy Optimization [53.81524570216593]
  Test-Time Policy Optimization (TTPO) is an asymmetric objective that distills agreeing rollouts via OPSD and penalizes disagreeing rollouts with Grouped RL.<n>We show that TTPO matches label-supervised OPSD on five competition-level benchmarks, raises Qwen3-1.7B from 38.0% to 45.2% in TTT, yields +25.2% to +36.4% without thinking, and shows strong cross-task generalization.
  arXiv  Detail & Related papers  (2026-08-27T17:58:10Z)
• Verifier-Free RL for LLMs via Intrinsic Gradient-Norm Reward [69.99652051809737]
  In this work, we propose Verifier-free Intrinsic Gradient-Norm Reward (VIGOR)<n>VIGOR is a simple reward that uses only the policy model itself.<n>It exhibits cross-domain transfer to code benchmarks when trained only on math data.
  arXiv  Detail & Related papers  (2026-05-11T03:15:37Z)
• Can LLMs Learn to Reason Robustly under Noisy Supervision? [31.39996460068968]
  We take the first step toward a systematic analysis of noisy label mechanisms in RLVR.<n>We distinguish two types of noise: inactive noisy labels, which reduce data efficiency, and active noisy labels, which are reinforced and risk skewing the model.<n>Motivated by this dynamic, we propose Online Label Refinement (OLR), which progressively corrects potentially noisy labels with majority-voted answers.
  arXiv  Detail & Related papers  (2026-04-05T06:30:50Z)
• What If Consensus Lies? Selective-Complementary Reinforcement Learning at Test Time [57.533031432715084]
  Test-Time Reinforcement Learning (TTRL) enables Large Language Models (LLMs) to enhance reasoning capabilities on unlabeled test streams.<n>Existing TTRL methods rely exclusively on positive pseudo-labeling strategies.<n>We propose SCRL, a robust test-time reinforcement learning framework that effectively mitigates label noise amplification.
  arXiv  Detail & Related papers  (2026-03-20T11:47:12Z)
• Tool Verification for Test-Time Reinforcement Learning [70.09740926883818]
  Test-time reinforcement learning (TTRL) has emerged as a promising paradigm for self-evolving large reasoning models.<n>We present T3RL (Tool-Verification for Test-Time Reinforcement Learning), which introduces test-time tool verification into reward estimation.
  arXiv  Detail & Related papers  (2026-03-02T18:57:52Z)
• Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning [17.384089089363382]
  We identify a root cause that existing methods overlook: the uniform penalization of errors.<n>Current approaches treat all incorrect rollouts within a group identically.<n>We propose the Asymmetric Confidence-aware Error Penalty (ACE)
  arXiv  Detail & Related papers  (2026-02-24T22:46:43Z)
• CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal [84.71254539482369]
  Group-relative reinforcement learning with verifiable rewards (RLVR) often wastes the most informative data it already has the failures.<n>We present CARE, a failure-centric post-training framework for multimodal reasoning that turns errors into supervision.<n> CARE improves accuracy and training smoothness while explicitly increasing the share of learning signal that comes from failures.
  arXiv  Detail & Related papers  (2025-12-22T16:34:21Z)
• ASPO: Asymmetric Importance Sampling Policy Optimization [31.38346888572171]
  The Importance Sampling (IS) ratios of positive-advantage tokens are mismatched, leading to unbalanced token weighting for positive and negative tokens.<n>This mismatch suppresses the update of low-probability tokens while over-amplifying already high-probability ones.<n>We propose Asymmetric Importance Sampling Policy Optimization (ASPO), which uses a simple yet effective strategy that flips the IS ratios of positive-advantage tokens.
  arXiv  Detail & Related papers  (2025-10-07T15:54:24Z)
• Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers [90.50039419576807]
  Reinforcement Learning with Verifiable Rewards (RLVR) trains policies against automated verifiers to avoid costly human labeling.<n>To reduce vulnerability to verifier hacking, many RLVR systems collapse rewards to binary $0,1$ during training.<n>This choice carries a cost: it introduces textitfalse negatives (rejecting correct answers, FNs) and textitfalse positives (accepting incorrect ones, FPs)
  arXiv  Detail & Related papers  (2025-10-01T13:56:44Z)
• Critique to Verify: Accurate and Honest Test-Time Scaling with RL-Trained Verifiers [63.99316853136304]
  Mirror-Critique is a framework that trains a verifier with informative critiques.<n>We deploy a small instruction-tuned model to synthesize high-quality critique data.<n>The resulting Mirror-Verifier is deployed to evaluate candidate solutions by generating multiple critiques per solution.
  arXiv  Detail & Related papers  (2025-09-27T06:50:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.