Graph Adversarial Diffusion Convolution
This work addresses robustness and performance issues in graph neural networks for tasks like graph signal denoising, though it appears incremental as it builds upon existing Graph Diffusion Convolution methods.
The paper tackles the Graph Signal Denoising problem by proposing a min-max optimization approach that introduces perturbations to the graph structure, resulting in the Graph Adversarial Diffusion Convolution (GADC) variant, which enhances robustness against adversarial attacks and improves performance on heterophilic graphs.
This paper introduces a min-max optimization formulation for the Graph Signal Denoising (GSD) problem. In this formulation, we first maximize the second term of GSD by introducing perturbations to the graph structure based on Laplacian distance and then minimize the overall loss of the GSD. By solving the min-max optimization problem, we derive a new variant of the Graph Diffusion Convolution (GDC) architecture, called Graph Adversarial Diffusion Convolution (GADC). GADC differs from GDC by incorporating an additional term that enhances robustness against adversarial attacks on the graph structure and noise in node features. Moreover, GADC improves the performance of GDC on heterophilic graphs. Extensive experiments demonstrate the effectiveness of GADC across various datasets. Code is available at https://github.com/SongtaoLiu0823/GADC.