Logical Distillation of Graph Neural Networks
This provides an interpretable alternative for graph learning tasks, though it is incremental as it builds on known connections between GNNs and logic.
The authors tackled the problem of making Graph Neural Networks (GNNs) interpretable by distilling them into logical models, resulting in distilled models that are interpretable, succinct, and achieve similar accuracy to the underlying GNNs, with performance exceeding GNNs when the ground truth is expressible in a specific logic fragment.
We present a logic based interpretable model for learning on graphs and an algorithm to distill this model from a Graph Neural Network (GNN). Recent results have shown connections between the expressivity of GNNs and the two-variable fragment of first-order logic with counting quantifiers (C2). We introduce a decision-tree based model which leverages an extension of C2 to distill interpretable logical classifiers from GNNs. We test our approach on multiple GNN architectures. The distilled models are interpretable, succinct, and attain similar accuracy to the underlying GNN. Furthermore, when the ground truth is expressible in C2, our approach outperforms the GNN.