Deep Graph Infomax

TL;DR

Deep Graph Infomax (DGI) introduces an unsupervised method for learning node representations in graphs by maximizing mutual information, outperforming some supervised methods on node classification tasks.

Contribution

DGI is a novel unsupervised approach that leverages mutual information maximization with GCNs, applicable to both transductive and inductive learning without relying on random walk objectives.

Findings

Competitive node classification performance

Outperforms some supervised methods

Applicable to transductive and inductive settings

Abstract

We present Deep Graph Infomax (DGI), a general approach for learning node representations within graph-structured data in an unsupervised manner. DGI relies on maximizing mutual information between patch representations and corresponding high-level summaries of graphs---both derived using established graph convolutional network architectures. The learnt patch representations summarize subgraphs centered around nodes of interest, and can thus be reused for downstream node-wise learning tasks. In contrast to most prior approaches to unsupervised learning with GCNs, DGI does not rely on random walk objectives, and is readily applicable to both transductive and inductive learning setups. We demonstrate competitive performance on a variety of node classification benchmarks, which at times even exceeds the performance of supervised learning.