Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data

• URL: http://arxiv.org/abs/2604.18493v1
• Date: Mon, 20 Apr 2026 16:43:28 GMT
• Title: Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data
• Abstract summary: We propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration.<n>We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance.<n> Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO.
• Score: 55.84428098924793
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reinforcement Learning (RL) enhances LLM reasoning, yet a paradox emerges as models scale: strong base models saturate standard benchmarks (e.g., MATH), yielding correct but homogeneous solutions. In such environments, the lack of failure cases causes the advantage signal in group-relative algorithms (e.g., GRPO) to vanish, driving policies into mode collapse. To address this, we propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration. Unlike standard sampling that follows model biases, CUTS flattens the local optimization landscape by sampling uniformly from constrained high-confidence candidates. We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance. Experiments on Qwen3 models demonstrate that our approach prevents policy degeneration and significantly boosts out-of-domain generalization. Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO, validating that maintaining diversity within the semantic manifold is critical for rigorous reasoning.

Related papers

• N-GRPO: Embedding-Level Neighbor Mixing for Enhanced Policy Optimization [55.14402862283128]
  We introduce N-GRPO, a novel exploration strategy integrated into the Group Relative Policy Optimization framework.<n>Rather than relying on token-level sampling or native embedding-level noise, our approach leverages Semantic Neighbor Mixing.<n>N-GRPO achieves consistent improvements over strong baselines on math reasoning benchmarks and also exhibits robust generalization capabilities on out-of-distribution tasks.
  arXiv  Detail & Related papers  (2026-06-09T12:21:27Z)
• Embedding-perturbed Exploration Preference Optimization for Flow Models [23.146684617048674]
  We propose a novel framework that sustains optimization through embedding-level perturbation.<n>Our method introduces structured, embedding-level perturbations within sample groups, guaranteeing a robust variance.<n>Our approach significantly outperforms state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-05-15T09:56:40Z)
• Policy-based Tuning of Autoregressive Image Models with Instance- and Distribution-Level Rewards [16.135177543347773]
  Autoregressive (AR) models are highly effective for image generation, yet their standard maximum-likelihood estimation training lacks direct optimization for sample quality and diversity.<n>We propose a lightweight RL framework that casts token-based AR inference as a Markov Decision Process, optimized via Group Relative Policy Optimization.<n>Our core contribution is the introduction of a novel distribution-level Leave-One-Out FID (LOO-FID) reward.
  arXiv  Detail & Related papers  (2026-03-24T11:28:36Z)
• When Right Meets Wrong: Bilateral Context Conditioning with Reward-Confidence Correction for GRPO [18.988527161000203]
  Group Relative Policy Optimization (GRPO) has emerged as an effective method for training reasoning models.<n>We present a contrastive reformulation of GRPO, showing that the GRPO objective implicitly maximizes the margin between the policy ratios of correct and incorrect samples.<n>We propose Bilateral Context Conditioning (BICC), a mechanism that allows the model to cross-reference successful and failed reasoning traces.
  arXiv  Detail & Related papers  (2026-03-13T16:25:02Z)
• Improving Search Agent with One Line of Code [68.58667107354253]
  Tool-based Agentic Reinforcement Learning (TARL) has emerged as a promising paradigm for training search agents to interact with external tools.<n>We propose textbfSearch textbfAgent textbfPolicy textbfOptimization (textbfSAPO), which stabilizes training via a conditional token-level KL constraint.
  arXiv  Detail & Related papers  (2026-03-10T04:07:39Z)
• Back to Basics: Revisiting Exploration in Reinforcement Learning for LLM Reasoning via Generative Probabilities [10.235183326885794]
  Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as an indispensable paradigm for enhancing reasoning in Large Language Models (LLMs)<n>We analyze this issue from the perspective of sampling probability dynamics, identifying that the standard objective disproportionately reinforces the highest-likelihood paths.<n>We propose a novel Advantage Re-weighting Mechanism (ARM) designed to equilibrate the confidence levels across all correct responses.
  arXiv  Detail & Related papers  (2026-02-05T04:06:55Z)
• Not All Preferences Are Created Equal: Stability-Aware and Gradient-Efficient Alignment for Reasoning Models [52.48582333951919]
  We propose a dynamic framework designed to enhance alignment reliability by maximizing the Signal-to-Noise Ratio of policy updates.<n>SAGE (Stability-Aware Gradient Efficiency) integrates a coarse-grained curriculum mechanism that refreshes candidate pools based on model competence.<n> Experiments on multiple mathematical reasoning benchmarks demonstrate that SAGE significantly accelerates convergence and outperforms static baselines.
  arXiv  Detail & Related papers  (2026-02-01T12:56:10Z)
• Group Distributionally Robust Optimization-Driven Reinforcement Learning for LLM Reasoning [45.86058898829962]
  Multi-Ad Distributionally Robust Optimization (GDRO) is an optimization-first framework that moves beyond uniform reasoning.<n>We propose two independent GDRO games for post-training: Prompt-GDRO, which employs an EMA-debiased multiplicative-weight bandit sampler to target the intensive difficulty margin and upweight persistently hard groups without frequency bias; and Rollout-GDRO, which uses a shadow-price controller to reallocate rollouts across groups, maximizing gradient variance reduction on hard tasks under a fixed mean budget (compute-neutral)<n>We validate our framework on the DAPO 14.1k dataset using Q
  arXiv  Detail & Related papers  (2026-01-27T07:10:41Z)
• CoT-Saliency: Unified Chain-of-Thought Reasoning for Heterogeneous Saliency Tasks [96.64597365827046]
  We present the first unified framework that jointly handles three operationally heterogeneous saliency tasks.<n>We introduce a Chain-of-Thought (CoT) reasoning process in a Vision-Language Model (VLM) to bridge task heterogeneity.<n>We show our model matches or outperforms specialized SOTA methods and strong closed-source VLMs across all tasks.
  arXiv  Detail & Related papers  (2025-11-01T04:37:01Z)
• Stratified GRPO: Handling Structural Heterogeneity in Reinforcement Learning of LLM Search Agents [90.45197506653341]
  Large language model (LLM) agents increasingly rely on external tools such as search engines to solve complex, multi-step problems.<n>The trajectories of search agents are structurally heterogeneous, where variations in the number, placement, and outcomes of search calls lead to fundamentally different answer directions and reward distributions.<n>Standard policy gradient methods, which use a single global baseline, suffer from what we identify and formalize as cross-stratum bias.<n>We propose Stratified GRPO, whose central component, Stratified Advantage Normalization (SAN), partitions trajectories into homogeneous strata based on their structural properties and computes advantages locally within each
  arXiv  Detail & Related papers  (2025-10-07T17:59:13Z)
• MMR1: Enhancing Multimodal Reasoning with Variance-Aware Sampling and Open Resources [113.33902847941941]
  Variance-Aware Sampling (VAS) is a data selection strategy guided by Variance Promotion Score (VPS)<n>We release large-scale, carefully curated resources containing 1.6M long CoT cold-start data and 15k RL QA pairs.<n> Experiments across mathematical reasoning benchmarks demonstrate the effectiveness of both the curated data and the proposed VAS.
  arXiv  Detail & Related papers  (2025-09-25T14:58:29Z)
• Single-Trajectory Distributionally Robust Reinforcement Learning [21.955807398493334]
  We propose Distributionally Robust RL (DRRL) to enhance performance across a range of environments. Existing DRRL algorithms are either model-based or fail to learn from a single sample trajectory. We design a first fully model-free DRRL algorithm, called distributionally robust Q-learning with single trajectory (DRQ)
  arXiv  Detail & Related papers  (2023-01-27T14:08:09Z)

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.