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

2026-01-12

AI Summary

This work addresses the suboptimal reasoning patterns that large reasoning models often exhibit on mathematical and scientific tasks due to training biases, which constrain their performance. The authors propose a reinforcement learning framework that, for the first time, enables dynamic selection of the optimal reasoning strategy on a per-problem basis. Their approach generates multiple candidate reasoning paths through multimodal rollback, employs a verifier to guide the selection of the best-performing mode, and incorporates attention masking to prevent mode leakage, thereby internalizing strategy selection as part of the model’s policy. Trained with GRPO, the method yields significant and consistent performance gains across multiple models and benchmarks, effectively mitigating suboptimal reasoning and enhancing both robustness and adaptability.

Technology Category

• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• reasoning patterns
• pattern sub-optimality
• large reasoning models
• problem-dependent reasoning
• reasoning strategy selection

Innovation

• Group Pattern Selection Optimization
• reasoning patterns
• reinforcement learning
• verifier-guided selection
• attention masking