From "Aha Moments" to Controllable Thinking: Toward Meta-Cognitive Reasoning in Large Reasoning Models via Decoupled Reasoning and Control

TL;DR

This paper introduces MERA, a framework that decouples reasoning and control in large reasoning models, enabling explicit regulation of reasoning processes to improve efficiency and accuracy.

Contribution

MERA is a novel framework that separates reasoning and control, utilizing auxiliary models and policy optimization to regulate reasoning in large models.

Findings

Improves reasoning efficiency and reduces latency.

Enhances reasoning accuracy on benchmarks.

Provides explicit reasoning-control traces.

Abstract

Large Reasoning Models (LRMs) have demonstrated a latent capacity for complex reasoning by spontaneously exhibiting cognitive behaviors such as step-by-step reasoning, reflection, and backtracking, commonly referred to as "Aha Moments". However, such emergent behaviors remain unregulated and uncontrolled, often resulting in overthinking, where the model continues generating redundant reasoning content even after reaching reliable conclusions. This leads to excessive computational costs and increased latency, limiting the practical deployment of LRMs. The root cause lies in the absence of intrinsic regulatory mechanisms, as current models are unable to monitor and adaptively manage their reasoning process to determine when to continue, backtrack, or terminate. To address this issue, we propose the Meta-cognitive Reasoning Framework (MERA), which explicitly decouples the thinking process into distinct reasoning and control components, thereby enabling the independent optimization of control strategies. Specifically, MERA incorporates a takeover-based data construction mechanism that identifies critical decision points during reasoning and delegates the creation of control signals to auxiliary LLMs, thereby enabling the construction of high-quality reasoning-control data. Additionally, a structured reasoning-control separation is implemented via supervised fine-tuning, enabling the model to generate explicit traces and acquire initial meta-cognitive control capabilities. Finally, MERA employs Control-Segment Policy Optimization (CSPO), which combines segment-wise Group Relative Policy Optimization (GRPO) with a control-masking mechanism to optimize control behavior learning while minimizing interference from irrelevant content. Experiments on various reasoning benchmarks demonstrate that models trained with MERA enhance both reasoning efficiency and accuracy.