Optimal Propagation for Graph Neural Networks

TL;DR

This paper introduces a bi-level optimization method to learn optimal graph structures for GNNs by directly learning the propagation matrix, improving robustness and efficiency in semi-supervised node classification.

Contribution

It proposes a novel bi-level optimization framework for jointly learning the graph structure and node classification, including a low-rank approximation for efficiency.

Findings

Outperforms baseline methods in accuracy and robustness

Demonstrates effectiveness on real-world datasets

Reduces computational complexity with low-rank approximation

Abstract

Graph Neural Networks (GNNs) have achieved tremendous success in a variety of real-world applications by relying on the fixed graph data as input. However, the initial input graph might not be optimal in terms of specific downstream tasks, because of information scarcity, noise, adversarial attacks, or discrepancies between the distribution in graph topology, features, and groundtruth labels. In this paper, we propose a bi-level optimization approach for learning the optimal graph structure via directly learning the Personalized PageRank propagation matrix as well as the downstream semi-supervised node classification simultaneously. We also explore a low-rank approximation model for further reducing the time complexity. Empirical evaluations show the superior efficacy and robustness of the proposed model over all baseline methods.