Hierarchical graph representation learning with differentiable pooling

Zhitao Ying, Jiaxuan You, Christopher Morris, Xiang Ren, William L. Hamilton, Jure Leskovec

Neural Information Processing Systems · 2018 · 909 citations · 0 references

Concepts

TL;DR

Graph neural networks have achieved state‑of‑the‑art node classification and link prediction, yet their flat architectures lack hierarchical representations, limiting performance on graph‑level classification tasks. This work introduces DiffPool, a differentiable pooling module that produces hierarchical graph representations and can be integrated end‑to‑end with any GNN architecture. DiffPool learns a soft cluster assignment for nodes at each GNN layer, collapsing nodes into clusters that serve as the coarsened input for the next layer. Experiments show that adding DiffPool to existing GNNs boosts graph‑classification accuracy by 5–10 % and sets a new state‑of‑the‑art on four of five benchmark datasets.

Abstract

Recently, graph neural networks (GNNs) have revolutionized the field of graph representation learning through effectively learned node embeddings, and achieved state-of-the-art results in tasks such as node classification and link prediction. However, current GNN methods are inherently flat and do not learn hierarchical representations of graphs---a limitation that is especially problematic for the task of graph classification, where the goal is to predict the label associated with an entire graph. Here we propose DiffPool, a differentiable graph pooling module that can generate hierarchical representations of graphs and can be combined with various graph neural network architectures in an end-to-end fashion. DiffPool learns a differentiable soft cluster assignment for nodes at each layer of a deep GNN, mapping nodes to a set of clusters, which then form the coarsened input for the next GNN layer. Our experimental results show that combining existing GNN methods with DiffPool yields an average improvement of 5-10% accuracy on graph classification benchmarks, compared to all existing pooling approaches, achieving a new state-of-the-art on four out of five benchmark datasets.