This is a past event.

Dr. Susan Janiszewski, MTRI

TITLE: Graph Convolutional Neural Networks and the Graph Laplacian

ABSTRACT: Convolutional Neural Networks (CNNs) have unlocked several capabilities related to object identification and classification, especially in the field of image processing.  The relatively new Graph Convolutional Neural Network is a generalization of the traditional image processing CNN, which allows for many of the techniques developed for images to be applied in an analogous way to graphs.  This talk will introduce GraphCNNs, discuss the link between GraphCNNs and traditional CNNs, and present hypotheses on how we can further improve GraphCNN performance by exploiting the graph's spectral properties through the graph Laplacian.