When Sharpening Becomes Collapse: Sampling Bias and Semantic Coupling in RL with Verifiable Rewards

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Fan, Mingyuan, Han, Weiguang, Wang, Daixin, Chen, Cen, Zhang, Zhiqiang, Zhou, Jun
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Machine Learning Computation and Language
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Towards High Data Efficiency in Reinforcement Learning with Verifiable Reward
di: Tang, Xinyu, et al.
Pubblicazione: (2025)

RubricEM: Meta-RL with Rubric-guided Policy Decomposition beyond Verifiable Rewards
di: Li, Gaotang, et al.
Pubblicazione: (2026)

Evaluating Interactive Reasoning in Large Language Models: A Hierarchical Benchmark with Executable Games
di: Fan, Mingyuan, et al.
Pubblicazione: (2026)

Diagnosing and Mitigating System Bias in Self-Rewarding RL
di: Tan, Chuyi, et al.
Pubblicazione: (2025)

Chart-RL: Generalized Chart Comprehension via Reinforcement Learning with Verifiable Rewards
di: Zhang, Xin, et al.
Pubblicazione: (2026)

Mitigating Lost in Multi-turn Conversation via Curriculum RL with Verifiable Accuracy and Abstention Rewards
di: Li, Ming, et al.
Pubblicazione: (2025)

Blockwise Advantage Estimation for Multi-Objective RL with Verifiable Rewards
di: Pavlenko, Kirill, et al.
Pubblicazione: (2026)

RAZOR: Sharpening Knowledge by Cutting Bias with Unsupervised Text Rewriting
di: Yang, Shuo, et al.
Pubblicazione: (2024)

FedMCP: Parameter-Efficient Federated Learning with Model-Contrastive Personalization
di: Zhao, Qianyi, et al.
Pubblicazione: (2024)

What Does the Server See? Understanding Privacy Leakage from Large Language Models in Split Inference
di: Fan, Mingyuan, et al.
Pubblicazione: (2026)

When Self-Belief Misleads: Active Label Acquisition for Reinforcement Learning with Verifiable Rewards
di: Wang, Li, et al.
Pubblicazione: (2026)

AGR: Age Group fairness Reward for Bias Mitigation in LLMs
di: Cao, Shuirong, et al.
Pubblicazione: (2024)

RewardFlow: Topology-Aware Reward Propagation on State Graphs for Agentic RL with Large Language Models
di: Feng, Xiao, et al.
Pubblicazione: (2026)

On-Policy RL with Optimal Reward Baseline
di: Hao, Yaru, et al.
Pubblicazione: (2025)

When Distance Distracts: Representation Distance Bias in BT-Loss for Reward Models
di: Xie, Tong, et al.
Pubblicazione: (2025)

Self-Consistency via Marginal Sharpening
di: Arzhantsev, Aleksei, et al.
Pubblicazione: (2026)

Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward
di: Huang, Guanhua, et al.
Pubblicazione: (2025)

On the Trustworthiness Landscape of State-of-the-art Generative Models: A Survey and Outlook
di: Fan, Mingyuan, et al.
Pubblicazione: (2023)

FlowRL: Matching Reward Distributions for LLM Reasoning
di: Zhu, Xuekai, et al.
Pubblicazione: (2025)

KodCode: A Diverse, Challenging, and Verifiable Synthetic Dataset for Coding
di: Xu, Zhangchen, et al.
Pubblicazione: (2025)

DLM-One: Diffusion Language Models for One-Step Sequence Generation
di: Chen, Tianqi, et al.
Pubblicazione: (2025)

TDRM: Smooth Reward Models with Temporal Difference for LLM RL and Inference
di: Zhang, Dan, et al.
Pubblicazione: (2025)

Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning
di: Setlur, Amrith, et al.
Pubblicazione: (2024)

DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards
di: Zhang, Kaiyi, et al.
Pubblicazione: (2026)

RL Tango: Reinforcing Generator and Verifier Together for Language Reasoning
di: Zha, Kaiwen, et al.
Pubblicazione: (2025)

BeamPERL: Parameter-Efficient RL with Verifiable Rewards Specializes Compact LLMs for Structured Beam Mechanics Reasoning
di: Hage, Tarjei Paule, et al.
Pubblicazione: (2026)

Offline RL by Reward-Weighted Fine-Tuning for Conversation Optimization
di: Mukherjee, Subhojyoti, et al.
Pubblicazione: (2025)

Self-Improvement in Language Models: The Sharpening Mechanism
di: Huang, Audrey, et al.
Pubblicazione: (2024)

ToolRL: Reward is All Tool Learning Needs
di: Qian, Cheng, et al.
Pubblicazione: (2025)

vCache: Verified Semantic Prompt Caching
di: Schroeder, Luis Gaspar, et al.
Pubblicazione: (2025)

Rethinking Sample Polarity in Reinforcement Learning with Verifiable Rewards
di: Tang, Xinyu, et al.
Pubblicazione: (2025)

CoVerRL: Breaking the Consensus Trap in Label-Free Reasoning via Generator-Verifier Co-Evolution
di: Pan, Teng, et al.
Pubblicazione: (2026)

DISA: Offline Importance Sampling for Distribution-Matching LLM-RL
di: Wang, Shaobo, et al.
Pubblicazione: (2026)

Rewarding the Rare: Uniqueness-Aware RL for Creative Problem Solving in LLMs
di: Hu, Zhiyuan, et al.
Pubblicazione: (2026)

Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards
di: Ma, Zhengzhao, et al.
Pubblicazione: (2026)

Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training
di: Xu, Ran, et al.
Pubblicazione: (2026)

Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains
di: Gunjal, Anisha, et al.
Pubblicazione: (2025)

Hybrid Reinforcement: When Reward Is Sparse, It's Better to Be Dense
di: Tao, Leitian, et al.
Pubblicazione: (2025)

Cascade Reward Sampling for Efficient Decoding-Time Alignment
di: Li, Bolian, et al.
Pubblicazione: (2024)

Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax
di: Su, Zeli, et al.
Pubblicazione: (2026)