Flexible Diffusion for Graph Neural Networks

Songwei Zhao

Keywords: Graph Neural Network; Diffusion; Smoothing Label

Abstract: Graph Neural Networks (GNNs) are attracting growing attention due to their promising performance in modeling a variety of graph-structured data. However, most existing GNNs only consider fixed-range discrete message passing and aggregation, none of them are aware of the importance of the degree and local structure of nodes for smoothing features, which significantly limits the applicability of GNNs. Furthermore, previous approaches either focus on adaptive selection for aggregation structures or treat discrete graph convolution as a continuous diffusion process, lacking holistic consideration of the two issues and resulting in the performance of the model being significantly limited. To this end, we propose a novel Flexible Diffusion Convolution (Flexi-DC), which aims to smooth features by setting a specific continuous diffusion for each node through the degree-and-local structure of the nodes. Specifically, Flexi-DC first extracts the degree and local structure knowledge of the nodes in the graph data and then injects it into the diffusion convolution module to smooth features. Additionally, we also utilize the extracted knowledge for smoothing labels. Flexi-DC is an efficient framework that can significantly improve the performance of most GNN architectures. Experimental results demonstrate that Flexi-DC outperforms their vanilla implementations by an average accuracy of 10.82\% (GCN), 12.33\% (JKNet), and 11.04\% (ARMA) on nine graph datasets with different homophily ratios.

Primary Area: learning on graphs and other geometries & topologies

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8759