Learning Graph Neural Networks with Noisy Labels

TL;DR

This paper investigates how to make Graph Neural Networks more robust to noisy labels by combining message-passing models with loss correction techniques, improving accuracy in noisy graph classification tasks.

Contribution

It introduces a noise-tolerant training approach for GNNs that enhances robustness to symmetric label noise using loss correction methods.

Findings

Test accuracy improves under symmetric noisy conditions.

Combining message-passing models with loss correction enhances robustness.

Approach is effective for graph classification with noisy labels.

Abstract

We study the robustness to symmetric label noise of GNNs training procedures. By combining the nonlinear neural message-passing models (e.g. Graph Isomorphism Networks, GraphSAGE, etc.) with loss correction methods, we present a noise-tolerant approach for the graph classification task. Our experiments show that test accuracy can be improved under the artificial symmetric noisy setting.