arXiv cs.AI· Yang Sheng, Jie Fu·· 2 天前AI 评分22
Circuit Claims Depend on What Is Extracted and How It Is Compared：电路提取结论取决于提取对象与比较方式
Circuit Claims Depend on What Is Extracted and How It Is Compared
AI 导读
一项在合成 Lean 策略预测基准上的研究发现，电路提取的结论取决于报告的是哪种电路以及如何提取和比较。同一 Transformer 的稠密与权重稀疏检查点间，组件连接重叠度低、有时接近随机基线；而 RL 在组合证明上的最大准确率提升，伴随组合电路中最多的组件落在对应原子电路之外，消融这些额外组件会降低组合准确率。作者据此提出电路提取研究的报告规范。
正文 · 原文
Abstract:Circuit extraction identifies model components whose presence preserves a target behavior under ablation, but preserving behavior does not single out one circuit. We study this in a synthetic Lean tactic-prediction benchmark with fixed proof rules and randomized surface form. Across dense and weight-sparse checkpoints of the same transformer, evaluated on atomic and compositional proofs, we vary which extracted object is reported: a compact prediction-preserving circuit, a broader graph retaining surrounding read, write, and routing structure, or a subgraph extracted with a post-ablation loss budget. First, overlap between groups of component-to-component connections is low, sometimes close to a random baseline. Overlap is much higher for selected attention-head sets under deterministic selection, but falls sharply when selection among equally scored heads is randomized. Second, the normalized graph-size ranking of conditions that differ in which supervised checkpoint initializes reinforcement learning (RL) agrees between repeated extractions of the compact circuit and between joint and separate query/key representations of each of the other two graphs. For the loss-budgeted graph, query/key ordering agreement persists through the budget sweep. Finally, across those same conditions, the largest exact-match accuracy gains from RL on compositional proofs come with the largest proportion of components in the compositional circuit that lie outside the corresponding atomic circuits. Ablating these additional components in the sparse RL model reduces compositional accuracy while largely preserving accuracy on the corresponding atomic tasks. A circuit-level claim therefore requires specifying which circuit is reported and how it is extracted and compared; behavior alone does not determine its interpretation. We distill these requirements into a reporting practice for circuit-extraction studies.
| Comments: | 35 pages. Accepted at NeurIPS 2026. Extended version with appendices |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.18921 [cs.LG] |
| (or arXiv:2607.18921v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2607.18921 arXiv-issued DOI via DataCite |
Submission history
From: Yang Sheng [view email]
[v1]
Tue, 21 Jul 2026 10:04:29 UTC (888 KB)
[v2]
Thu, 15 Oct 2026 09:58:16 UTC (1,149 KB)
来源：arXiv cs.AI · arxiv.org