Globally Interpretable Graph Learning via Distribution Matching

Yi Nian, Yurui Chang, Wei Jin, Lu Lin

Keywords: Globally Interpretable Graph Learning via Distribution Matching

Abstract:

Graphs neural networks (GNNs) have emerged as a powerful graph learning model due to their superior capacity in capturing critical graph patterns. Instead of treating GNNs as black boxes in an end-to-end fashion of training and deployment, people start to turn their attention to understand and explain the basis of model behavior. Existing works mainly focus on local interpretation, which aims to reveal the discriminative pattern for individual instances. However, the retrieved pattern cannot be directly generalized to reflect the high-level model behavior, i.e., patterns captured by the model for a certain class. To gain global insights about graph learning mechanism, we aim to answer an important question that is not yet studied: em how to provide a global interpretation for the model training procedure? We formulate this problem as globally interpretable graph learning, which targets on distilling high-level and human-intelligible patterns that dominate the learning procedure, such that training on this pattern can recover a similar model.

To address this problem, we first propose a new interpretation metric tailored for evaluating the fidelity of the resulting model. Our preliminary analysis shows that interpretative patterns generated by existing global methods fail to recover the model training procedure. Thus, we further propose our solution, Graph Distribution Matching (GDM), which synthesizes interpretive graphs by matching the distribution of the original and interpretive graphs in the feature space of the GNN as its training proceeds. These few interpretive graphs demonstrate the most informative patterns the model captures during training. Extensive experiments on graph classification datasets demonstrate multiple advantages of the proposed method, including high model fidelity, predictive accuracy and time efficiency, as well as the ability to reveal class-relevant structure.

Track: Graph Algorithms and Learning for the Web

Submission Guidelines Scope: Yes

Submission Guidelines Blind: Yes

Submission Guidelines Format: Yes

Submission Guidelines Limit: Yes

Submission Guidelines Authorship: Yes

Student Author: Yes

Submission Number: 2146