发表机构
University of Kentucky; Dalhousie University（肯塔基大学; 达尔豪斯大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出GAPS方法，通过维度级门控补充现有条件激活引导方法，在毒性缓解与概念移除任务上，提升了语言模型的行为-能力权衡表现。
AI 中文摘要
激活引导通过在生成过程中向隐藏状态添加引导向量，来抑制语言模型中不合需要的行为。近期的条件方法如CAST和DSAS，通过决定何时进行干预，改善了行为与能力的权衡，但它们一旦激活，就会将完整的密集向量应用于所有隐藏维度，无论神经元是否携带概念信息，或是否已处于期望的状态。我们引入维度级条件作为补充的选择性轴，还决定要对哪些神经元进行干预。我们的方法GAPS（基于后验与可分性的门控激活引导）结合了两种无需训练的门控：静态可分性门控，它通过AUROC将引导限制在具有统计可靠概念信息的神经元上；以及动态后验门控，它仅在神经元当前激活能被高斯模型下的不合需要概念更好解释时，才对该神经元进行引导。这些门控每个令牌增加O(D)的开销，且可插入现有条件方法中。在使用Gemma-3（4B）和Qwen-3（1.7B）进行的毒性缓解（RealToxicityPrompts）和概念移除（OneSeC）任务上，GAPS始终与令牌级对应方法的帕累托前沿相匹配或有所提升；在固定能力预算下，DSAS+GAPS将Gemma-3的毒性率从6.52%降至0.48%，而仅使用DSAS时为3.52%。消融实验表明，大部分增益来自后验门控。
英文摘要
Activation steering suppresses undesired behaviors in language models by adding a steering vector to the hidden state during generation. Recent conditional methods such as CAST and DSAS improve the behavior-capability trade-off by deciding when to intervene, but once active, they apply the full dense vector to all hidden dimensions, regardless of whether a neuron carries concept information or already lies in the desired regime. We introduce dimension-level conditioning as a complementary axis of selectivity that also decides which neurons to intervene on. Our method, GAPS (Gated Activation steering via Posterior and Separability), combines two training-free gates: a static separability gate that restricts steering to neurons with statistically reliable concept information (via AUROC), and a dynamic posterior gate that steers a neuron only when its current activation is better explained by the undesired concept under a Gaussian model. The gates add O(D) overhead per token, and they plug into existing conditional methods. On toxicity mitigation (RealToxicityPrompts) and concept removal (OneSeC) with Gemma-3 (4B) and Qwen-3 (1.7B), GAPS consistently matches or improves the Pareto front of its token-level counterparts; under a fixed capability budget, DSAS+GAPS reduces Gemma-3's toxicity rate from 6.52% to 0.48%, versus 3.52% for DSAS alone. Ablations attribute most of the gain to the posterior gate.
