analiza aiqbench
This paper proposes Graph Circuit Learning (GCL), a framework using graph neural networks to generalize circuit localization across multiple transformer model-task pairs without retraining for each specific instance. By framing mechanistic interpretability as a graph machine learning problem, the authors achieve high edge AUROC scores on held-out cases, suggesting that GNNs can effectively model interactions among computational pathways in transformers. This work bridges the gap between general graph ML techniques and the specialized field of mechanistic interpretability.
Can Graph Learning Learn Circuits?
arXiv:2608.08536v2 Announce Type: replace Abstract: Circuit localization is a mechanistic interpretability task whose goal is to identify a sparse subgraph of a transformer's computation graph sufficient to reproduce a particular behavior. Most established methods localize circuits independently for each model-task pair. We instead frame circuit localization as a graph machine learning problem in which the edges of a computation graph represent computational pathways, and graph neural networks (GNNs) model interactions among these pathways. We introduce Graph Circuit Learning (GCL), a supervised, amortized framework that trains a GNN across multiple model-task pairs and applies it to unseen cases. To provide sufficient data, we augment the InterpBench benchmark with additional cases derived from TracrBench programs. Of the 46 evaluated GCL configurations, the one selected by cross-validation scored a median edge AUROC of 0.946 (interquartile interval [0.841, 0.980]) on the 16 held-out cases of our evaluation split. Removing all message-passing edges reduces the median to 0.926. We also adapt PGExplainer, a GNN explainability method, to circuit localization, obtaining a median edge AUROC of 0.950 on the same cases. These preliminary results suggest that graph machine learning offers a natural and potentially powerful perspective on circuit localization, one we hope encourages closer exchange between the two communities.
Title:Can Graph Learning Learn Circuits?
Abstract:Circuit localization is a mechanistic interpretability task whose goal is to identify a sparse subgraph of a transformer's computation graph sufficient to reproduce a particular behavior. Most established methods localize circuits independently for each model-task pair. We instead frame circuit localization as a graph machine learning problem in which the edges of a computation graph represent computational pathways, and graph neural networks (GNNs) model interactions among these pathways. We introduce Graph Circuit Learning (GCL), a supervised, amortized framework that trains a GNN across multiple model-task pairs and applies it to unseen cases. To provide sufficient data, we augment the InterpBench benchmark with additional cases derived from TracrBench programs. Of the 46 evaluated GCL configurations, the one selected by cross-validation scored a median edge AUROC of 0.946 (interquartile interval [0.841, 0.980]) on the 16 held-out cases of our evaluation split. Removing all message-passing edges reduces the median to 0.926. We also adapt PGExplainer, a GNN explainability method, to circuit localization, obtaining a median edge AUROC of 0.950 on the same cases. These preliminary results suggest that graph machine learning offers a natural and potentially powerful perspective on circuit localization, one we hope encourages closer exchange between the two communities.
Machine Learning (cs.LG)
arXiv:2608.08536 [cs.LG]
(or arXiv:2608.08536v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.08536 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 295 cuvinte
Subiecte conexe
Graph neural network predicts qubit routing costs
Quantum Zeitgeist · 18 sept. 2026
TRACE: Learning to Compute on Circuit Graphs
arXiv · 17 iun. 2026
Inductive Graph Representation Learning with Quantum Graph Neural Networks
arXiv · 22 sept. 2026
Variational Quantum Attention for Molecular Graph Learning
arXiv · 7 oct. 2026