Diffusion-Convolutional Neural Networks

TL;DR

Diffusion-Convolutional Neural Networks (DCNNs) introduce a diffusion-based operation for graph data, enabling effective node classification with invariant representations and efficient GPU implementation, outperforming existing relational models.

Contribution

The paper proposes a novel diffusion-convolution operation for graph neural networks, providing a new way to learn invariant representations for graph-structured data.

Findings

DCNNs outperform probabilistic relational models in node classification.

DCNNs are efficiently implementable on GPUs using tensor operations.

The model provides invariant representations under graph isomorphism.

Abstract

We present diffusion-convolutional neural networks (DCNNs), a new model for graph-structured data. Through the introduction of a diffusion-convolution operation, we show how diffusion-based representations can be learned from graph-structured data and used as an effective basis for node classification. DCNNs have several attractive qualities, including a latent representation for graphical data that is invariant under isomorphism, as well as polynomial-time prediction and learning that can be represented as tensor operations and efficiently implemented on the GPU. Through several experiments with real structured datasets, we demonstrate that DCNNs are able to outperform probabilistic relational models and kernel-on-graph methods at relational node classification tasks.