Scalable Consistency Training for Graph Neural Networks via Self-Ensemble Self-Distillation

TL;DR

This paper introduces a novel consistency training method for large-scale graph neural networks that leverages neighbor subsampling randomness to improve accuracy, especially in low-label scenarios.

Contribution

It proposes a self-ensemble self-distillation approach that approximates expected predictions over neighbor samples, enhancing GNN performance on large graphs.

Findings

Outperforms standard GNN training methods.

Achieves significant gains with low label rates.

Requires only a few samples for effective training.

Abstract

Consistency training is a popular method to improve deep learning models in computer vision and natural language processing. Graph neural networks (GNNs) have achieved remarkable performance in a variety of network science learning tasks, but to date no work has studied the effect of consistency training on large-scale graph problems. GNNs scale to large graphs by minibatch training and subsample node neighbors to deal with high degree nodes. We utilize the randomness inherent in the subsampling of neighbors and introduce a novel consistency training method to improve accuracy. For a target node we generate different neighborhood expansions, and distill the knowledge of the average of the predictions to the GNN. Our method approximates the expected prediction of the possible neighborhood samples and practically only requires a few samples. We demonstrate that our training method outperforms standard GNN training in several different settings, and yields the largest gains when label rates are low.