Polynomial-based graph convolutional neural networks for graph classification

Graph convolutional neural networks exploit convolution operators, based on some neighborhood aggregating scheme, to compute representations of graphs. The most common operators only local topological information. To consider wider receptive fields, the mainstream approach is non-linearly stack multiple graph (GC) layers. In this way, however, interactions among GC parameters at different levels pose a bias flow paper, we propose strategy, considering single layer that independently exploits neighbouring nodes distances, generating decoupled for each them. These are then processed by subsequent readout We implement strategy introducing polynomial (PGC) layer, prove being more expressive than and their linear stacking. Our contribution not limited definition operator with larger field, but both theoretically experimentally way non-linear convolutions stacked limits network expressiveness. Specifically, show architecture PGC achieves state art performance many commonly adopted classification benchmarks.