arXiv cs.CL· Jipeng Li, Pramod Khargonekar, Yanning Shen·· 11 小时前AI 评分31
通过受控符号输入因果定位 Transformer 中的证据效应
Causal Localization of Evidence Effects in Transformers
AI 导读
研究提出受控证据聚合（CEA）与平衡效用因果选择（BUC），通过改变提示中两个符号的计数与顺序，定位影响 Transformer 决策的注意力头。在 Pythia-410M 中，按四种顺序条件各选出的 1-12 个注意力头被消融后，可消除 65.5-91.6% 的计数效应。
正文 · 原文
Abstract:Can controlled symbolic inputs reveal attention heads that influence a language model's decisions from natural language evidence? We introduce controlled evidence aggregation (CEA), which varies the counts and order of two symbols in otherwise fixed prompts and measures changes in a frozen Transformer's preference between two answer tokens. The preference is measured by their logit difference; the prompts prescribe no correct answer. Balanced utility causal selection (BUC) identifies compact head sets by jointly evaluating how much ablating their outputs removes the count effect and how closely patching mean outputs between count conditions recovers the effect in both directions. We derive conditional bounds linking head attribution to intervention effects and specifying when the attribution shortlist can be recovered from finite samples. In Pythia-410M, ablating sets of 1-12 heads selected separately for four ordering regimes removes 65.5-91.6% of the count effect. The sets retain their influence on count comparisons not used for selection. To test relevance beyond symbolic inputs, we evaluate a natural language task asking which state is supported by more reports. Ablating eight heads selected only on symbolic yes/no prompts lowers mean accuracy by 8.2 percentage points, versus 0.8 for random sets matched by size and layer. Our results demonstrate that controlled symbolic probes can identify compact attention head sets whose causal influence extends to natural language decisions.
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.36524 [cs.CL] |
| (or arXiv:2610.36524v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36524 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jipeng Li [view email]
[v1]
Tue, 27 Oct 2026 02:27:36 UTC (288 KB)
来源：arXiv cs.CL · arxiv.org