Certified Circuits: Stability Guarantees for Mechanistic Circuits

2026-02-26

AI Summary

This work addresses the instability of existing mechanistic interpretability methods, whose discovered circuits are highly sensitive to training data and often fail out-of-distribution, potentially capturing dataset artifacts rather than genuine concepts. The paper introduces the first provably stable circuit discovery framework, which wraps any black-box circuit discovery algorithm with randomized data subsampling and edit-distance-based model perturbations, while pruning unstable neurons based on statistical confidence. This approach provides formal stability guarantees for circuit components, ensuring that explanations align with the intended target concept. Evaluated on ImageNet and out-of-distribution data, the certified circuits achieve up to a 91% improvement in accuracy and use 45% fewer neurons, maintaining reliability even when baseline methods fail.

Technology Category

• Machine Learning: Probabilistic Circuits and Graphical Models
• Computer Vision: Interpretability, Explainability, and Transparency
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

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 concept or 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 circuit component inclusion decisions are invariant to bounded edit-distance perturbations of the concept dataset. Unstable neurons are abstained from, yielding circuits that are more compact and more accurate. On ImageNet and OOD datasets, certified circuits achieve up to 91% higher accuracy while using 45% fewer neurons, 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 will be released soon!

Problem

• mechanistic interpretability
• circuit discovery
• stability
• out-of-distribution
• concept dataset

Innovation

• Certified Circuits
• mechanistic interpretability
• stability guarantees
• circuit discovery
• out-of-distribution robustness