Apr, 2024

增强带通滤波器逼近的光谱图神经网络的提升

TL;DRSpectral Graph Neural Networks (GNNs) have been widely used for graph learning tasks, but existing poly-GNNs face issues in approximating band-pass graph filters. In this paper, a novel poly-GNN called TrigoNet is proposed, which constructs graph filters using trigonometric polynomials and achieves superior performance in approximating band-pass graph filters.