Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering

2026-08-31

AI Summary

本文通过黎曼优化学习参数高效的旋转变换来控制大语言模型的拒绝行为，无需辅助结构，实验证明了该方法的有效性。

Abstract

Activation steering has emerged as a lightweight approach for controlling model refusal at inference time. A growing line of research explores trainable rotations of activations to develop geometrically principled intervention mechanisms. However, existing techniques rely on auxiliary constructs, such as refusal vectors, to define these rotations. In our work, we develop a self-contained methodology for learning parameter-efficient rotational transformations based on Riemannian optimization. We empirically validate the proposed scheme, demonstrating its superiority in intervention efficiency. An extensive ablation study highlights the importance of key design choices in our method. Our results identify the proposed rotation-based steering scheme as a promising direction for more reliable control over the behavior of LLMs.

Problem

• Activation Steering
• Refusal Behavior
• Riemannian Optimization
• Large Language Models (LLMs)

Innovation

• Riemannian optimization
• Activation steering
• Parameter-efficient rotations
• LLM control