Graph matching as a graph convolution operator for graph neural networks

Convolutional neural networks (CNNs), in a few decades, have outperformed the existing state of art methods classification context. However, way they were formalised, CNNs are bound to operate on euclidean spaces. Indeed, convolution is signal operation that defined This has restricted deep learning main use euclidean-defined data such as sound or image. And yet, numerous computer application fields (among which network analysis, computational social science, chemo-informatics graphics) induce non-euclideanly graphs, manifolds. In this paper we propose new architecture, directly into graph space. Convolution and pooling operators domain thanks matching procedure between input filter. We show its usability back-propagation Experimental results our model performance at level simple tasks. It shows robustness with respect changes improvement other non-euclidean convolutional architectures.