Sparse Diffusion-Convolutional Neural Networks

TL;DR

This paper introduces a thresholding method for diffusion-convolutional neural networks that significantly reduces memory usage from quadratic to linear in the number of nodes, enabling scalable node classification.

Contribution

The paper proposes a simple thresholding technique that reduces the memory complexity of DCNNs from O(N^2) to O(N) without compromising accuracy.

Findings

Memory complexity reduced from O(N^2) to O(N)

Predictive performance remains stable after thresholding

Method enables scalable node classification on large graphs

Abstract

The predictive power and overall computational efficiency of Diffusion-convolutional neural networks make them an attractive choice for node classification tasks. However, a naive dense-tensor-based implementation of DCNNs leads to $\mathcal{O}(N^2)$ memory complexity which is prohibitive for large graphs. In this paper, we introduce a simple method for thresholding input graphs that provably reduces memory requirements of DCNNs to O(N) (i.e. linear in the number of nodes in the input) without significantly affecting predictive performance.