03/01/19 A Review of Graph Convolutional Neural Networks

The first part of this workshop will be a review of neural networks in tensorflow and keras. The second part will go into an exciting specific type of neural network called graph convolutional neural networks. There are numerous real-world data in non-euclidean relations. Finding an optimum representation of these types of data can be useful to investigate their hidden patterns and structures. 2-d manifolds in a 3-d space and graph-embedded relations are two important examples of data points in a non-euclidean relation. Graph convolutional neural networks, as an emerging and surprisingly successful tool, can be used to capture these relations. We cover the mathematics behind this method and provide a survey on the most recent works on GCNs. We also dive into the implementation of one or two basic networks of these types.