Distance Metric Learning using Graph Convolutional Networks: Application to Functional Brain Networks

摘要

Evaluating similarity between graphs is of major importance in severalcomputer vision and pattern recognition problems, where graph representationsare often used to model objects or interactions between elements. The choice ofa distance or similarity metric is, however, not trivial and can be highlydependent on the application at hand. In this work, we propose a novel metriclearning method to evaluate distance between graphs that leverages the power ofconvolutional neural networks, while exploiting concepts from spectral graphtheory to allow these operations on irregular graphs. We demonstrate thepotential of our method in the field of connectomics, where neuronal pathwaysor functional connections between brain regions are commonly modelled asgraphs. In this problem, the definition of an appropriate graph similarityfunction is critical to unveil patterns of disruptions associated with certainbrain disorders. Experimental results on the ABIDE dataset show that our methodcan learn a graph similarity metric tailored for a clinical application,improving the performance of a simple k-nn classifier by 11.9% compared to atraditional distance metric.