arXiv cs.CL· Nishkal Hundia, Swastik Agrawal, Navita Goyal, Sarah Wiegreffe·· 11 小时前AI 评分31
少扫参数多引导：用激活引导几何减少超参数搜索
Sweep Less, Steer More: Using Activation Steering Geometry for Reduced Hyperparameter Search
AI 导读
研究基于可判别性指数 d' 这一计算廉价的几何指标，在 3 个模型的 5 种行为上验证其可用于替代逐层扫描来选择引导层。在多项选择问答（MCQA）中 d' 能可靠定位有效引导层，在开放式多 token 生成中同样适用于层选择。但用生成 token 表示在 DIM 线上的位置预测单例引导成败，仅在 d' 较高的数据集上有效，开放式生成中跨行为、模型和训练 token 选择并不一致。
正文 · 原文
Abstract:Representation steering methods like difference-in-means (DIM) offer a lightweight approach to controlling LLM behavior, but their effectiveness varies widely across concepts, datasets, and layers, making principled deployment costly without exhaustive hyperparameter sweeps. We build on the discriminability index $d'$, a compute-cheap geometric measure of how well contrastive training activations separate along the steering direction. Across five behaviors on 3 models, we demonstrate uses beyond its established role as a dataset-level steerability predictor. First, in the multiple-choice question answering (MCQA) setting, we show that $d'$ reliably identifies performant layers at which to steer, eliminating the need for a layer-wise sweep. Second, we extend this analysis to open-ended multi-token generation, and find that $d'$ remains successful for layer selection across different settings. Lastly, we test whether the position of a generated token's representation on the DIM line predicts steering success on that individual example. In the MCQA setting, this works for high-$d'$ datasets but is inconsistent across behaviors, models, and training-token choices in open-ended generation. This highlights that methods developed and conclusions drawn in single-token MCQA settings do not necessarily transfer to open-ended generation.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.01496 [cs.LG] |
| (or arXiv:2611.01496v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.01496 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Nishkal Hundia [view email]
[v1]
Sun, 1 Nov 2026 03:29:29 UTC (1,495 KB)
来源：arXiv cs.CL · arxiv.org