Timezone: »

Graph Information Bottleneck
Tailin Wu · Hongyu Ren · Pan Li · Jure Leskovec

Thu Dec 10 09:00 PM -- 11:00 PM (PST) @ Poster Session 6 #1836

Representation learning of graph-structured data is challenging because both graph structure and node features carry important information. Graph Neural Networks (GNNs) provide an expressive way to fuse information from network structure and node features. However, GNNs are prone to adversarial attacks. Here we introduce Graph Information Bottleneck (GIB), an information-theoretic principle that optimally balances expressiveness and robustness of the learned representation of graph-structured data. Inheriting from the general Information Bottleneck (IB), GIB aims to learn the minimal sufficient representation for a given task by maximizing the mutual information between the representation and the target, and simultaneously constraining the mutual information between the representation and the input data. Different from the general IB, GIB regularizes the structural as well as the feature information. We design two sampling algorithms for structural regularization and instantiate the GIB principle with two new models: GIB-Cat and GIB-Bern, and demonstrate the benefits by evaluating the resilience to adversarial attacks. We show that our proposed models are more robust than state-of-the-art graph defense models. GIB-based models empirically achieve up to 31% improvement with adversarial perturbation of the graph structure as well as node features.

Author Information

Tailin Wu (Stanford)
Tailin Wu

Tailin Wu is a postdoc researcher in the Department of Computer Science at Stanford University, working with professor Jure Leskovec. His research interests lies in AI for large-scale simulations of complex systems, AI for scientific discovery, and representation learning

Hongyu Ren (Stanford University)
Pan Li (Purdue University)
Jure Leskovec (Stanford University/Pinterest)

More from the Same Authors