Spectral-designed depthwise separable graph neural networks - Normandie Université Access content directly
Conference Papers Year :

Spectral-designed depthwise separable graph neural networks

Pierre Héroux
Benoit Gaüzère
Paul Honeine


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.
Fichier principal
Vignette du fichier
20.icml.depthwise.pdf (368.78 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03088372 , version 1 (26-12-2020)


  • HAL Id : hal-03088372 , version 1


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⟩
186 View
104 Download


Gmail Facebook Twitter LinkedIn More