Relational Pooling for Graph Representations

Abstract

This work generalizes graph neural networks (GNNs) beyond those based on the Weisfeiler-Lehman (WL) algorithm, graph Laplacians, and diffusions. Our approach, denoted Relational Pooling (RP), draws from the theory of finite partial exchangeability to provide a framework with maximal representation power for graphs. RP can work with existing graph representation models and, somewhat counterintuitively, can make them even more powerful than the original WL isomorphism test. Additionally, RP allows architectures like Recurrent Neural Networks and Convolutional Neural Networks to be used in a theoretically sound approach for graph classification. We demonstrate improved performance of RP-based graph representations over state-of-the-art methods on a number of tasks.

ICML 2019 Camera-Ready. Added to molecular experiments and balanced the classes of the validation folds for the synthetic-graph experiments. Clarified some discussions

References in corpus (10)

• Inductive Representation Learning on Large Graphs
• Geometric deep learning: going beyond Euclidean data
• Molecular Graph Convolutions: Moving Beyond Fingerprints
• Hierarchical Graph Representation Learning with Differentiable Pooling
• Pitfalls of Graph Neural Network Evaluation
• Group Equivariant Convolutional Networks
• Graph limits and exchangeable random graphs
• Equivariance Through Parameter-Sharing
• Deep Models of Interactions Across Sets
• On the Combinatorial Power of the Weisfeiler-Lehman Algorithm

Cited by in corpus (13)

• Principal Neighbourhood Aggregation for Graph Nets
• On the equivalence between graph isomorphism testing and function approximation with GNNs
• Can Graph Neural Networks Count Substructures?
• Provably Powerful Graph Networks
• Global Attention Improves Graph Networks Generalization
• Molecule Property Prediction and Classification with Graph Hypernetworks
• A pipeline for fair comparison of graph neural networks in node classification tasks
• Graph convolutions that can finally model local structure
• Towards Scale-Invariant Graph-related Problem Solving by Iterative Homogeneous Graph Neural Networks
• Breaking the Expressive Bottlenecks of Graph Neural Networks
• Deep Constraint-based Propagation in Graph Neural Networks
• Improving Expressivity of Graph Neural Networks
• A Collective Learning Framework to Boost GNN Expressiveness