Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Group Pattern Selection Optimization: Let LRMs Pick the Right Pattern for Reasoning

Hanbin Wang, Jingwei Song, Jinpeng Li, Fei Mi, Lifeng Shang

TL;DR

This paper tackles sub-optimal fixed reasoning patterns in large reasoning models by showing substantial performance variance across patterns on math and science benchmarks. It introduces Group Pattern Selection Optimization (GPSO), a reinforcement learning framework that extends RLVR/GRPO with multi-pattern rollouts, verifier-guided pattern selection, and attention-masking to learn per-instance optimal reasoning strategies without leaking exploration scaffolds. GPSO demonstrates consistent gains across multiple model backbones and benchmarks (AIME 2024/2025, GPQA, MATH-500), including notable improvements on challenging tasks and robust generalization. The work provides a practical, model-agnostic approach to adaptive reasoning and supplies open-source code and data to facilitate reproducibility.

Abstract

Large reasoning models (LRMs) exhibit diverse high-level reasoning patterns (e.g., direct solution, reflection-and-verification, and exploring multiple solutions), yet prevailing training recipes implicitly bias models toward a limited set of dominant patterns. Through a systematic analysis, we identify substantial accuracy variance across these patterns on mathematics and science benchmarks, revealing that a model's default reasoning pattern is often sub-optimal for a given problem. To address this, we introduce Group Pattern Selection Optimization (GPSO), a reinforcement learning framework that extends GRPO by incorporating multi-pattern rollouts, verifier-guided optimal pattern selection per problem, and attention masking during optimization to prevent the leakage of explicit pattern suffixes into the learned policy. By exploring a portfolio of diverse reasoning strategies and optimizing the policy on the most effective ones, GPSO enables the model to internalize the mapping from problem characteristics to optimal reasoning patterns. Extensive experiments demonstrate that GPSO delivers consistent and substantial performance gains across various model backbones and benchmarks, effectively mitigating pattern sub-optimality and fostering more robust, adaptable reasoning. All data and codes are available at https://github.com/wanghanbinpanda/GPSO.

Group Pattern Selection Optimization: Let LRMs Pick the Right Pattern for Reasoning

TL;DR

This paper tackles sub-optimal fixed reasoning patterns in large reasoning models by showing substantial performance variance across patterns on math and science benchmarks. It introduces Group Pattern Selection Optimization (GPSO), a reinforcement learning framework that extends RLVR/GRPO with multi-pattern rollouts, verifier-guided pattern selection, and attention-masking to learn per-instance optimal reasoning strategies without leaking exploration scaffolds. GPSO demonstrates consistent gains across multiple model backbones and benchmarks (AIME 2024/2025, GPQA, MATH-500), including notable improvements on challenging tasks and robust generalization. The work provides a practical, model-agnostic approach to adaptive reasoning and supplies open-source code and data to facilitate reproducibility.

Abstract

Large reasoning models (LRMs) exhibit diverse high-level reasoning patterns (e.g., direct solution, reflection-and-verification, and exploring multiple solutions), yet prevailing training recipes implicitly bias models toward a limited set of dominant patterns. Through a systematic analysis, we identify substantial accuracy variance across these patterns on mathematics and science benchmarks, revealing that a model's default reasoning pattern is often sub-optimal for a given problem. To address this, we introduce Group Pattern Selection Optimization (GPSO), a reinforcement learning framework that extends GRPO by incorporating multi-pattern rollouts, verifier-guided optimal pattern selection per problem, and attention masking during optimization to prevent the leakage of explicit pattern suffixes into the learned policy. By exploring a portfolio of diverse reasoning strategies and optimizing the policy on the most effective ones, GPSO enables the model to internalize the mapping from problem characteristics to optimal reasoning patterns. Extensive experiments demonstrate that GPSO delivers consistent and substantial performance gains across various model backbones and benchmarks, effectively mitigating pattern sub-optimality and fostering more robust, adaptable reasoning. All data and codes are available at https://github.com/wanghanbinpanda/GPSO.
Paper Structure (23 sections, 8 equations, 5 figures, 3 tables)

This paper contains 23 sections, 8 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Reinforcement Learning with Verifiable Rewards
  4. Sampling Strategies for Reinforcement Learning
  5. Reasoning Patterns of Large Reasoning Models
  6. Methodology
  7. Preliminaries of RLVR
  8. Group Pattern Selection Optimization (GPSO)
  9. Multi-Pattern Rollout.
  10. Pattern Selection Rule.
  11. Attention Masking for Pattern Suffix.
  12. Training Objective.
  13. Experimental Methodology
  14. Evaluation Results
  15. Overall Performance
  16. ...and 8 more sections

Figures (5)

  • Figure 1: Comparison of model performance under different reasoning patterns on three benchmarks (AIME2024, AIME2025, and GPQA): (a) DeepSeek-R1-0528 and (b) Qwen3-8B (Thinking). No: No reasoning prompt, Dir: Direct solution, Ref: Reflection and verification, Exp: Explore Multiple solutions, Best: Pattern selected with the highest accuracy on each question.
  • Figure 2: Overview of Group Pattern Selection Optimal (GPSO).
  • Figure 3: Training Accuracy Curves On AIME2024 and AIME2025.
  • Figure 4: Pattern Usage Distribution Before and After GPSO Training
  • Figure 5: Distribution of reasoning patterns used by various LLMs on MATH and Science tasks