NeuReasoner: Towards Explainable, Controllable, and Unified Reasoning via Mixture-of-Neurons

TL;DR

NeuReasoner introduces a white-box, neuron-based framework for explainable, controllable, and unified reasoning in large models, addressing multiple failure modes and improving performance across benchmarks.

Contribution

It identifies key neurons linked to reasoning failures and develops NeuReasoner, a framework that integrates failure detection and self-correction for enhanced reasoning.

Findings

Achieves up to 27.0% performance improvement.

Reduces token consumption by up to 63.3%.

Demonstrates effectiveness across six benchmarks and multiple models.

Abstract

Large Reasoning Models (LRMs) have recently achieved remarkable success in complex reasoning tasks. However, closer scrutiny reveals persistent failure modes compromising performance and cost: I) Intra-step level, marked by calculation or derivation errors; II) Inter-step level, involving oscillation and stagnation; and III) Instance level, causing maladaptive over-thinking. Existing endeavors target isolated levels without unification, while their black-box nature and reliance on RL hinder explainability and controllability. To bridge these gaps, we conduct an in-depth white-box analysis, identifying key neurons (Mixture of Neurons, MoN) and their fluctuation patterns associated with distinct failures. Building upon these insights, we propose NeuReasoner, an explainable, controllable, and unified reasoning framework driven by MoN. Technically, NeuReasoner integrates lightweight MLPs for failure detection with a special token-triggered self-correction mechanism learned via SFT. During inference, special tokens are inserted upon failure detection to actuate controllable remedial behaviors. Extensive evaluations across six benchmarks, six backbone models (8B~70B) against nine competitive baselines, demonstrate that NeuReasoner achieves performance gains of up to 27.0% while reducing token consumption by 19.6% ~ 63.3%.