Article Source

• Performance Adaptive Sampling Strategy Towards Fast and Accurate Graph Neural Networks

Performance Adaptive Sampling Strategy Towards Fast and Accurate Graph Neural Networks

Abstract

The scalability issue of GNNs mainly comes from uncontrollable neighborhood expansion in the aggregation stage. The only way to alleviate this neighbor explosion problem is to sample a fixed number of neighbors in the aggregation operation. In this paper, we propose a performance-adaptive sampling strategy PASS that samples neighbors informative for a target task. PASS outperforms state-of-the-art sampling methods by up to 10% accuracy on public benchmarks and up to 53% accuracy in the presence of adversarial attacks.