跳到正文
arXiv cs.AI· Nima Eshraghi, Lovedeep Gondara, Sagarika Suresh, Xiaotong Xu, Jithin Pradeep, Fanny Chevalier·· 4 小时前AI 评分27

从逐 Token 门控到轨迹感知激活引导:CFP 与 CLABS 两种门控方法解析

Right Token, Right Time: From Per-Token Gating to Trajectory-Aware Activation Steering

AI 导读

一项 arXiv 研究提出基于稀疏自编码器(SAE)的激活引导新框架,将引导信号视为固定可引导 token 预算下的逐 token 分配问题。研究引入两种门控:无记忆的对比特征投影(CFP)和轨迹感知的闭环自适应伯努利引导(CLABS),后者将分配建模为闭环反馈控制。在同等预算下,两者实现最大的行为偏移,CLABS 在紧预算下表现最强且最好地保持流畅性。

正文 · 原文

View PDF HTML (experimental)

Abstract:Activation steering via sparse autoencoders (SAEs) enables behavioral control of large language models without task-specific fine-tuning. Yet, standard methods apply the steering signal at every generated token, displacing representations from the distribution the model was trained on and eroding generation quality. We show that \emph{when} along a generation the intervention is applied is a decisive axis of control. We cast steering as a per-token allocation problem under a fixed budget of steerable tokens, so that methods differ only in which tokens they select, making allocation the sole variable. Within this framework we introduce two gates. The first, Contrastive Feature Projection (CFP), is memoryless and selects the tokens least aligned with the target behavior through a sparse, interpretable feature projection. The second, Closed-Loop Adaptive Bernoulli Steering (CLABS), is trajectory aware. It treats allocation as closed-loop feedback control, tracking the alignment a generation has accumulated so far and spending budget where the trajectory falls behind. On a linear steering-response model we derive an explicit gain region that guarantees closed-loop stability of CLABS and show that its integral term reduces the steady-state tracking offset. Empirically, CFP and CLABS achieve the largest behavioral shift among budgeted gates at matched budgets, with CLABS strongest under the tight budgets where allocation matters most, while best preserving fluency.
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.18274 [cs.LG]
  (or arXiv:2611.18274v1 [cs.LG] for this version)
  https://doi.org/10.48550/arXiv.2611.18274

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Nima Eshraghi Dr [view email]
[v1] Fri, 13 Nov 2026 15:06:43 UTC (101 KB)

来源:arXiv cs.AI · arxiv.org