search menu icon-carat-right cmu-wordmark

Graph Convolutional Neural Networks

October 2019 Poster
Oren Wright

This poster describes research to use graph convolutional neural networks to extend deep learning.

Publisher:

Software Engineering Institute

Subjects

Abstract

A growing number of DoD data problems are graph problems: the data from sources such as sensor feeds and web traffi c require graphs to represent mathematically. Machine learning seems like a perfect tool for such datasets, but machine learning approaches for the irregularly structured data of graph problems are sharply limited. We use graph signal processing formalisms to create new tools for graph convolutional neural networks (GCNNs), extending deep learning into the irregular world of graph problems.