Notice
This record has been edited as far as possible, missing data will be added when the version of record is issued.
Metadata only
Date
2024-05-10Type
- Conference Paper
ETH Bibliography
yes
Altmetrics
Abstract
We propose a new self-explainable Graph Neural Network (GNN) model: GraphChef. GraphChef integrates decision trees into the GNN message passing framework. Given a dataset, GraphChef returns a set of rules (a recipe) that explains each class in the dataset unlike existing GNNs and explanation methods that reason on individual graphs. Thanks to the decision trees, GraphChef recipes are human understandable. We also present a new pruning method to produce small and easy to digest trees. Experiments demonstrate that GraphChef reaches comparable accuracy to not self-explainable GNNs and produced decision trees are indeed small. We further validate the correctness of the discovered recipes on datasets where explanation ground truth is available: Reddit-Binary, MUTAG, BA-2Motifs, BA-Shapes, Tree-Cycle, and Tree-Grid. Show more
Publication status
publishedExternal links
Book title
The Twelfth International Conference on Learning RepresentationsPublisher
OpenReviewEvent
Organisational unit
03604 - Wattenhofer, Roger / Wattenhofer, Roger
Notes
Poster presentation on May 10, 2024.More
Show all metadata
ETH Bibliography
yes
Altmetrics