arXiv cs.CY· Alaa Anani, Tobias Lorenz, Bernt Schiele, Mario Fritz, Jonas Fischer·· 11 小时前AI 评分38
Certified Circuits：为机制可解释性电路提供稳定性保证
Certified Circuits: Stability Guarantees for Mechanistic Circuits
AI 导读
Certified Circuits 为电路发现提供可证明的稳定性保证，通过随机数据子采样包装任意黑盒发现算法，确保电路组件的纳入决策对概念数据集的有界编辑距离扰动保持不变。
正文 · 原文
Abstract:Understanding how neural networks arrive at their predictions is essential for debugging, auditing, and deployment. Mechanistic interpretability pursues this goal by identifying circuits--minimal subnetworks responsible for specific behaviors. However, existing circuit discovery methods are brittle: circuits depend strongly on the chosen concept dataset and often fail to transfer out-of-distribution, raising doubts whether they capture the concept or merely dataset-specific artifacts. We introduce Certified Circuits, which provide provable stability guarantees for circuit discovery. Our framework wraps any black-box discovery algorithm with randomized data subsampling to certify that inclusion decisions over circuit components--neurons or edges of the model graph, depending on the base algorithm--are invariant to bounded edit-distance perturbations of the concept dataset. Unstable components are abstained from, yielding circuits that are more compact and more accurate. We validate across three architectures (ResNet, ViT, GPT-2) on vision (ImageNet and four OOD datasets) and language (IOI, IOI-Hard, Greater-Than) tasks. Certified circuits achieve up to 56% higher accuracy and up to 80% fewer components, and remain reliable where baselines degrade. Certified Circuits puts circuit discovery on formal ground by producing mechanistic explanations that are provably stable and better aligned with the target concept. Code: this https URL.
| Comments: | Published at ICML 2026 |
| Subjects: | Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Computers and Society (cs.CY) |
| Cite as: | arXiv:2602.22968 [cs.AI] |
| (or arXiv:2602.22968v4 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2602.22968 arXiv-issued DOI via DataCite |
|
| Journal reference: | International Conference on Machine Learning (ICML), 2026 |
Submission history
From: Alaa Anani [view email]
[v1]
Thu, 26 Feb 2026 13:07:31 UTC (1,208 KB)
[v2]
Mon, 2 Mar 2026 13:21:23 UTC (1,208 KB)
[v3]
Thu, 28 May 2026 21:41:37 UTC (8,104 KB)
[v4]
Wed, 4 Nov 2026 13:27:41 UTC (8,101 KB)
来源：arXiv cs.CY · arxiv.org