search menu icon-carat-right cmu-wordmark

Graph Convolutional Neural Networks

This project used graph signal processing formalisms to create new deep learning tools for graph convolutional neural networks (GCNNs).

Software Engineering Institute


Our approach employed topology-adaptive graph convolutional networks, introduced in 2017 by researchers at Carnegie Mellon University. This new class of GCNNs is grounded in graph signal processing theory and can be applied to any graph topology without the limiting assumptions and approximations of other methods. Our goal was to produce practical tools for mission problems such as cybersecurity, infrastructure monitoring, social network monitoring, and U.S. Army Research Laboratory work on translational neuroscience.