Reducing oversmoothing through informed weight initialization in graph neural networks

Abstract

In this work, we generalize the ideas of Kaiming initialization to Graph Neural Networks (GNNs) and propose a new scheme (G-Init) that reduces oversmoothing, leading to very good results in node and graph classification tasks. GNNs are commonly initialized using methods designed for other types of Neural Networks, overlooking the underlying graph topology. We analyze theoretically the variance of signals flowing forward and gradients flowing backward in the class of convolutional GNNs. We then simplify our analysis to the case of the GCN and propose a new initialization method. Results indicate that the new method (G-Init) reduces oversmoothing in deep GNNs, facilitating their effective use. Our approach achieves an accuracy of 61.60% on the CS dataset (32-layer GCN) and 69.24% on Cora (64-layer GCN), surpassing state-of-the-art initialization methods by 25.6 and 8.6 percentage points, respectively. Extensive experiments confirm the robustness of our method across multiple benchmark datasets, highlighting its effectiveness in diverse settings. Furthermore, our experimental results support the theoretical findings, demonstrating the advantages of deep networks in scenarios with no feature information for unlabeled nodes (i.e., “cold start” scenario).