Spectral-designed depthwise separable graph neural networks

Muhammet Balcilar (1) , Guillaume Renton (1) , Pierre Héroux (1) , Benoit Gaüzère (1) , Sébastien Adam (1) , Paul Honeine (1)

1 DocApp - LITIS - Equipe Apprentissage

This paper aims at revisiting Convolutional Graph Neural Networks (ConvGNNs) by designing new graph convolutions in spectral domain with a custom frequency profile while applying them in the spatial domain. Within the proposed framework, we propose two ConvGNNs methods: one using a simple single-convolution kernel that operates as a low-pass filter, and one operating multiple convolution kernels called Depthwise Separable Graph Convolution Network (DSGCN). The latter is a generalization of the depthwise separable convolution framework for graph convolutional networks, which allows to decrease the total number of trainable parameters while keeping the capacity of the model unchanged. Our proposals are evaluated on both transductive and inductive graph learning problems, demonstrating that DSGCN outperforms the state-of-the-art methods.

Domains: Statistics [stat]; Machine Learning [stat.ML]; Engineering Sciences [physics]; Signal and Image processing; Mathematics [math]; Statistics [math.ST]; Computer Science [cs]; Signal and Image Processing; Computer Science [cs]; Neural and Evolutionary Computing [cs.NE]; Computer Science [cs]; Machine Learning [cs.LG]; Computer Science [cs]; Computers and Society [cs.CY]; Computer Science [cs]; Computer Vision and Pattern Recognition [cs.CV]; Computer Science [cs]; Artificial Intelligence [cs.AI]

Muhammet Balcilar, Guillaume Renton, Pierre Héroux, Benoit Gaüzère, Sébastien Adam, et al.. Spectral-designed depthwise separable graph neural networks. Proceedings of Thirty-seventh International Conference on Machine Learning (ICML 2020) - Workshop on Graph Representation Learning and Beyond (GRL+ 2020), Jul 2020, Vienna, Austria. ⟨hal-03088372⟩