MSRS: Adaptive Multi-Subspace Representation Steering for Attribute Alignment in Large Language Models

2025-08-14

AI Summary

Existing multi-attribute behavioral control methods for large language models suffer from interference and trade-offs due to attribute coupling. To address this, we propose the Multi-Subspace Representation Guidance (MSRG) framework, which decouples distinct control attributes via orthogonal subspaces. MSRG introduces three key mechanisms: (i) a shared-specialized subspace decomposition, (ii) token-level semantic-sensitive intervention, and (iii) dynamic weight fusion—enabling fine-grained, low-interference collaborative control. Crucially, it avoids explicit attribute conflict without requiring hand-crafted constraints. Empirically, MSRG significantly outperforms state-of-the-art methods on instruction following, style transfer, and other multi-attribute tasks, while demonstrating strong generalization to unseen downstream tasks. Its core innovation lies in formulating attribute control as differentiable, dynamic, token-level interventions within orthogonal subspaces—thereby achieving both effective disentanglement and compositional flexibility.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: Behavior Learning & Control
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Activation steering offers a promising approach to controlling the behavior of Large Language Models by directly manipulating their internal activations. However, most existing methods struggle to jointly steer multiple attributes, often resulting in interference and undesirable trade-offs. To address this challenge, we propose Multi-Subspace Representation Steering (MSRS), a novel framework for effective multi-attribute steering via subspace representation fine-tuning. MSRS reduces inter-attribute interference by allocating orthogonal subspaces to each attribute, isolating their influence within the model's representation space. MSRS also incorporates a hybrid subspace composition strategy: it combines attribute-specific subspaces for unique steering directions with a shared subspace for common steering directions. A dynamic weighting function learns to efficiently integrate these components for precise control. During inference, MSRS introduces a token-level steering mechanism that dynamically identifies and intervenes on the most semantically relevant tokens, enabling fine-grained behavioral modulation. Experimental results show that MSRS significantly reduces attribute conflicts, surpasses existing methods across a range of attributes, and generalizes effectively to diverse downstream tasks.

Problem

• Control multiple attributes in LLMs without interference
• Allocate orthogonal subspaces to isolate attribute influence
• Enable fine-grained behavioral modulation via token-level steering

Innovation

• Orthogonal subspaces reduce attribute interference
• Hybrid subspace composition combines specific and shared directions
• Token-level steering enables fine-grained behavioral modulation