Central Virginia Chapter Dinner and Presentation - August 18, 2021
I propose a novel way of classifying graph-based architectures using two machine learning techniques: Deep Graph Convolutional Neural Networks (DGCNN) and Graph Convolutional Networks (GCN) with EigenPooling. These unique algorithms can help improve the system engineering process by streamlining the computational processes required for architecture selection. Graph classification using graph neural networks has attracted increasing attention, and this study shows how it can be used within systems engineering and design. Neural networks typically deal with input and output data in tensor form, but recently, there have been significant breakthroughs using graphs, and since graph theory is being seen more and more not only in systems engineering but engineering as a whole, these techniques show a promising breakthrough. Some applications this can be applied to include, but are not limited to, electrical circuits to aircraft thermal management systems, aircraft bleed air systems, cyber networks, and any system that can be represented by a graph.