Salvato in:
Dettagli Bibliografici
Autori principali: Dong, Fangan, Yan, Zuming, Ge, Xuri, Xu, Zhiwei, Zhang, Mengqi, Chen, Xuanang, He, Ben, Xin, Xin, Chen, Zhumin, Zhou, Ying
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2601.19847
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908792213471232
author Dong, Fangan
Yan, Zuming
Ge, Xuri
Xu, Zhiwei
Zhang, Mengqi
Chen, Xuanang
He, Ben
Xin, Xin
Chen, Zhumin
Zhou, Ying
author_facet Dong, Fangan
Yan, Zuming
Ge, Xuri
Xu, Zhiwei
Zhang, Mengqi
Chen, Xuanang
He, Ben
Xin, Xin
Chen, Zhumin
Zhou, Ying
contents Despite the strong reasoning capabilities of recent large language models (LLMs), achieving reliable performance on challenging tasks often requires post-training or computationally expensive sampling strategies, limiting their practical efficiency. In this work, we first show that a small subset of neurons in LLMs exhibits strong predictive correlations with reasoning correctness. Based on this observation, we propose AdaRAS (Adaptive Reasoning Activation Steering), a lightweight test-time framework that improves reasoning reliability by selectively intervening on neuron activations. AdaRAS identifies Reasoning-Critical Neurons (RCNs) via a polarity-aware mean-difference criterion and adaptively steers their activations during inference, enhancing incorrect reasoning traces while avoiding degradation on already-correct cases. Experiments on 10 mathematics and coding benchmarks demonstrate consistent improvements, including over 13% gains on AIME-24 and AIME-25. Moreover, AdaRAS exhibits strong transferability across datasets and scalability to stronger models, outperforming post-training methods without additional training or sampling cost.
format Preprint
id arxiv_https___arxiv_org_abs_2601_19847
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Identifying and Transferring Reasoning-Critical Neurons: Improving LLM Inference Reliability via Activation Steering
Dong, Fangan
Yan, Zuming
Ge, Xuri
Xu, Zhiwei
Zhang, Mengqi
Chen, Xuanang
He, Ben
Xin, Xin
Chen, Zhumin
Zhou, Ying
Computation and Language
Despite the strong reasoning capabilities of recent large language models (LLMs), achieving reliable performance on challenging tasks often requires post-training or computationally expensive sampling strategies, limiting their practical efficiency. In this work, we first show that a small subset of neurons in LLMs exhibits strong predictive correlations with reasoning correctness. Based on this observation, we propose AdaRAS (Adaptive Reasoning Activation Steering), a lightweight test-time framework that improves reasoning reliability by selectively intervening on neuron activations. AdaRAS identifies Reasoning-Critical Neurons (RCNs) via a polarity-aware mean-difference criterion and adaptively steers their activations during inference, enhancing incorrect reasoning traces while avoiding degradation on already-correct cases. Experiments on 10 mathematics and coding benchmarks demonstrate consistent improvements, including over 13% gains on AIME-24 and AIME-25. Moreover, AdaRAS exhibits strong transferability across datasets and scalability to stronger models, outperforming post-training methods without additional training or sampling cost.
title Identifying and Transferring Reasoning-Critical Neurons: Improving LLM Inference Reliability via Activation Steering
topic Computation and Language
url https://arxiv.org/abs/2601.19847