Steer2Edit：从激活引导到组件级编辑
Steer2Edit: From Activation Steering to Component-Level Editing
AI总结：
Steer2Edit通过将引导信号转化为可解释的参数更新，实现组件级权重编辑，提升模型的安全性、真实性和推理效率。
摘要：
引导方法通过在隐藏表示中识别语义方向来影响大语言模型的行为，但通常是通过推理时的激活干预，将固定的全局修改应用到模型的内部状态上。虽然有效，但此类干预在强控制下往往导致不理想的属性-效用权衡，因为它们忽略了许多行为是由模型组件的小而异质的子集所支配的。我们提出了Steer2Edit，一个理论基础扎实、无需训练的框架，将推理时的引导向量从控制信号转换为组件级的秩1权重编辑诊断信号。与其在生成过程中均匀地注入一个引导方向，Steer2Edit有选择地将行为影响重新分布到单个注意力头和MLP神经元上，产生可解释的编辑，保持标准前向传递并保持与优化并行推理的兼容性。在安全对齐、幻觉缓解和推理效率方面，Steer2Edit始终实现了更优的属性-效用权衡：在匹配的下游性能下，它提高了安全性高达17.2%，增加了真实性9.8%，并平均减少了推理长度12.2%。总体而言，Steer2Edit通过将引导信号转换为可解释、无训练的参数更新，为表示引导和权重编辑之间提供了原则性的桥梁。我们的代码可在https://github.com/Trustworthy-ML-Lab/Steer2Edit上获取。
英文摘要：
Steering methods influence Large Language Model behavior by identifying semantic directions in hidden representations, but are typically realized through inference-time activation interventions that apply a fixed, global modification to the model's internal states. While effective, such interventions often induce unfavorable attribute-utility trade-offs under strong control, as they ignore the fact that many behaviors are governed by a small and heterogeneous subset of model components. We propose Steer2Edit, a theoretically grounded, training-free framework that transforms steering vectors from inference-time control signals into diagnostic signals for component-level rank-1 weight editing. Instead of uniformly injecting a steering direction during generation, Steer2Edit selectively redistributes behavioral influence across individual attention heads and MLP neurons, yielding interpretable edits that preserve the standard forward pass and remain compatible with optimized parallel inference. Across safety alignment, hallucination mitigation, and reasoning efficiency, Steer2Edit consistently achieves more favorable attribute-utility trade-offs: at matched downstream performance, it improves safety by up to 17.2%, increases truthfulness by 9.8%, and reduces reasoning length by 12.2% on average. Overall, Steer2Edit provides a principled bridge between representation steering and weight editing by translating steering signals into interpretable, training-free parameter updates. Our code is available at https://github.com/Trustworthy-ML-Lab/Steer2Edit
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
