Controllable LLM Reasoning via Sparse Autoencoder-Based Steering

TL;DR

This paper introduces SAE-Steering, a novel method using sparse autoencoders to control and improve reasoning strategies in large reasoning models, leading to more reliable and accurate reasoning paths.

Contribution

It proposes a two-stage feature identification pipeline with sparse autoencoders to effectively control reasoning strategies in LRMs, surpassing existing methods in control effectiveness.

Findings

SAE-Steering filters out over 99% of irrelevant features.

It achieves over 15% improvement in control effectiveness.

It improves reasoning accuracy by 7% by redirecting erroneous paths.

Abstract

Large Reasoning Models (LRMs) exhibit human-like cognitive reasoning strategies (e.g. backtracking, cross-verification) during reasoning process, which improves their performance on complex tasks. Currently, reasoning strategies are autonomously selected by LRMs themselves. However, such autonomous selection often produces inefficient or even erroneous reasoning paths. To make reasoning more reliable and flexible, it is important to develop methods for controlling reasoning strategies. Existing methods struggle to control fine-grained reasoning strategies due to conceptual entanglement in LRMs' hidden states. To address this, we leverage Sparse Autoencoders (SAEs) to decompose strategy-entangled hidden states into a disentangled feature space. To identify the few strategy-specific features from the vast pool of SAE features, we propose SAE-Steering, an efficient two-stage feature identification pipeline. SAE-Steering first recalls features that amplify the logits of strategy-specific keywords, filtering out over 99\% of features, and then ranks the remaining features by their control effectiveness. Using the identified strategy-specific features as control vectors, SAE-Steering outperforms existing methods by over 15\% in control effectiveness. Furthermore, controlling reasoning strategies can redirect LRMs from erroneous paths to correct ones, achieving a 7\% absolute accuracy improvement.