Learning Graph While Training: An Evolving Graph Convolutional Neural Network
This addresses a bottleneck in graph-based machine learning for domains like chemistry where graph structures are irregular, offering a more flexible approach.
The paper tackles the problem of applying graph convolutional networks to data with diverse or undefined graph structures by proposing an evolving graph convolutional network (EGCN) that learns graph Laplacians during training. The result is superior performance with accelerated parameter fitting and significantly improved prediction accuracy on multiple datasets.
Convolution Neural Networks on Graphs are important generalization and extension of classical CNNs. While previous works generally assumed that the graph structures of samples are regular with unified dimensions, in many applications, they are highly diverse or even not well defined. Under some circumstances, e.g. chemical molecular data, clustering or coarsening for simplifying the graphs is hard to be justified chemically. In this paper, we propose a more general and flexible graph convolution network (EGCN) fed by batch of arbitrarily shaped data together with their evolving graph Laplacians trained in supervised fashion. Extensive experiments have been conducted to demonstrate the superior performance in terms of both the acceleration of parameter fitting and the significantly improved prediction accuracy on multiple graph-structured datasets.