arXiv cs.CL· Haoyan Luo, Mateo Espinosa Zarlenga, Mateja Jamnik·· 10 小时前AI 评分31
SKOP:通过键正交投影实现激活引导,减少 LLM 推理与检索性能退化
Don't Lose Focus: Activation Steering via Key-Orthogonal Projections
AI 导读
研究者提出激活引导方法 SKOP(Steering via Key-Orthogonal Projections),通过保留模型依赖的少量 focus token 上的注意力模式、仅允许尾部 token 间重新分配,来抑制引导向量造成的注意力重路由。
正文 · 原文
Abstract:Activation steering controls LLM behaviour towards target behaviour by intervening in internal representations, yet it often degrades reasoning and retrieval performance. We argue that a primary cause of this trade-off is attention rerouting: steering vectors alter query-key matching, shifting attention away from contextually important tokens toward less informative ones. To address this, we propose Steering via Key-Orthogonal Projections (SKOP), a steering method that constrains harmful attention rerouting without eliminating steering efficacy. SKOP achieves this by preserving attention patterns on a small set of focus tokens the model relies on for reasoning and retrieval, while allowing redistribution among less critical tail tokens. Across multiple steering benchmarks, we show that SKOP achieves the best joint steering-utility trade-off, reducing utility degradation by 5-7x while retaining over 95% of vanilla steering efficacy. Our results further suggest that, in long-context retrieval settings where vanilla steering approaches are ineffective, SKOP can maintain robust performance by avoiding attention rerouting.
| Comments: | NeurIPS 2026 spotlight |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.06342 [cs.CL] |
| (or arXiv:2605.06342v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.06342 arXiv-issued DOI via DataCite |
Submission history
From: Haoyan Luo [view email]
[v1]
Thu, 7 May 2026 14:29:18 UTC (6,229 KB)
[v2]
Thu, 19 Nov 2026 18:43:57 UTC (3,769 KB)
来源:arXiv cs.CL · arxiv.org