AI Insight
This paper introduces Certified Circuits, a method for identifying stable and reliable subnetworks within neural networks that are responsible for specific behaviors. The approach uses randomized data subsampling to provide mathematical guarantees that discovered circuits remain valid even when the training data is perturbed, addressing a key weakness in existing mechanistic interpretability methods that often fail when tested on out-of-distribution data. Testing across vision and language models (ResNet, ViT, GPT-2), the method produces circuits with up to 56% higher accuracy and 80% fewer components compared to baseline approaches.
Why it matters
This work addresses a critical challenge in AI safety and interpretability by ensuring that explanations of neural network behavior are robust rather than dataset-specific artifacts. The ability to certify circuit stability could improve trust in AI systems for high-stakes applications like healthcare and autonomous systems, where understanding and validating model decisions is essential for safe deployment.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
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: https://github.com/AlaaAnani/certified-circuits.
Source: Certified Circuits: Stability Guarantees for Mechanistic Circuits