Cover Image

When and Why Does Unsupervised RL Succeed in Mathematical Reasoning? A Manifold Envelopment Perspective

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhang, Zelin, Cheng, Fei, Chu, Chenhui
Format:	Preprint
Published:	2026
Subjects:	Machine Learning Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Predicting Where Steering Vectors Succeed
by: Billa, Jayadev
Published: (2026)

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?
by: Kim, Jeonghye, et al.
Published: (2026)

Scaling Up Membership Inference: When and How Attacks Succeed on Large Language Models
by: Puerto, Haritz, et al.
Published: (2024)

RL Grokking Recipe: How Does RL Unlock and Transfer New Algorithms in LLMs?
by: Sun, Yiyou, et al.
Published: (2025)

Why Reasoning Matters? A Survey of Advancements in Multimodal Reasoning (v1)
by: Bi, Jing, et al.
Published: (2025)

An Investigation of Robustness of LLMs in Mathematical Reasoning: Benchmarking with Mathematically-Equivalent Transformation of Advanced Mathematical Problems
by: Hao, Yuren, et al.
Published: (2025)

When Sharpening Becomes Collapse: Sampling Bias and Semantic Coupling in RL with Verifiable Rewards
by: Fan, Mingyuan, et al.
Published: (2026)

Improving RL Exploration for LLM Reasoning through Retrospective Replay
by: Dou, Shihan, et al.
Published: (2025)

Why Code, Why Now: An Information-Theoretic Perspective on the Limits of Machine Learning
by: Zhao, Zhimin
Published: (2026)

REA-RL: Reflection-Aware Online Reinforcement Learning for Efficient Reasoning
by: Deng, Hexuan, et al.
Published: (2025)

Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical Reasoning
by: Ding, Bowen, et al.
Published: (2025)

Exploring the Limit of Outcome Reward for Learning Mathematical Reasoning
by: Lyu, Chengqi, et al.
Published: (2025)

TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning
by: Wu, Jinyang, et al.
Published: (2025)

Schoenfeld's Anatomy of Mathematical Reasoning by Language Models
by: Li, Ming, et al.
Published: (2025)

When Does Divide and Conquer Work for Long Context LLM? A Noise Decomposition Framework
by: Xu, Zhen, et al.
Published: (2025)

Attention as a Compass: Efficient Exploration for Process-Supervised RL in Reasoning Models
by: Liu, Runze, et al.
Published: (2025)

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

Attention Meets Post-hoc Interpretability: A Mathematical Perspective
by: Lopardo, Gianluigi, et al.
Published: (2024)

A Symbolic Framework for Evaluating Mathematical Reasoning and Generalisation with Transformers
by: Meadows, Jordan, et al.
Published: (2023)

Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning
by: Liu, Zihe, et al.
Published: (2025)

AdaCuRL: Adaptive Curriculum Reinforcement Learning with Invalid Sample Mitigation and Historical Revisiting
by: Li, Renda, et al.
Published: (2025)

Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information
by: Shen, Guobin, et al.
Published: (2026)

Unlocking Multimodal Mathematical Reasoning via Process Reward Model
by: Luo, Ruilin, et al.
Published: (2025)

DRA-GRPO: Your GRPO Needs to Know Diverse Reasoning Paths for Mathematical Reasoning
by: Chen, Xiwen, et al.
Published: (2025)

Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Process
by: Zhang, Zhenyu, et al.
Published: (2025)

A Decomposition Perspective to Long-context Reasoning for LLMs
by: Xiao, Yanling, et al.
Published: (2026)

Improve Mathematical Reasoning in Language Models by Automated Process Supervision
by: Luo, Liangchen, et al.
Published: (2024)

Decomposing Elements of Problem Solving: What "Math" Does RL Teach?
by: Qin, Tian, et al.
Published: (2025)

When Does Multimodality Lead to Better Time Series Forecasting?
by: Zhang, Xiyuan, et al.
Published: (2025)

CFSafety: Comprehensive Fine-grained Safety Assessment for LLMs
by: Liu, Zhihao, et al.
Published: (2024)

Genius: A Generalizable and Purely Unsupervised Self-Training Framework For Advanced Reasoning
by: Xu, Fangzhi, et al.
Published: (2025)

Mitigating Overthinking in Large Reasoning Models via Manifold Steering
by: Huang, Yao, et al.
Published: (2025)

Embedding Trajectory for Out-of-Distribution Detection in Mathematical Reasoning
by: Wang, Yiming, et al.
Published: (2024)

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

We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning?
by: Qiao, Runqi, et al.
Published: (2024)

Learning to Reason as Action Abstractions with Scalable Mid-Training RL
by: Zhang, Shenao, et al.
Published: (2025)

Self-Error-Instruct: Generalizing from Errors for LLMs Mathematical Reasoning
by: Yu, Erxin, et al.
Published: (2025)

RL on Incorrect Synthetic Data Scales the Efficiency of LLM Math Reasoning by Eight-Fold
by: Setlur, Amrith, et al.
Published: (2024)

Offline Exploration-Aware Fine-Tuning for Long-Chain Mathematical Reasoning
by: Mu, Yongyu, et al.
Published: (2026)

ProcessBench: Identifying Process Errors in Mathematical Reasoning
by: Zheng, Chujie, et al.
Published: (2024)