Learning Graph Dynamics using Deep Neural Networks

A large number of real-world problems have high dimensional data. The data obtained from these problems is highly structured and usually in the form of graphs. Graphs represent spatial information about the system in the form of vertices and edges. Often graphs evolve with time and the underlying system exhibits dynamic behavior. Hence, these graphs contain both spatial and temporal information about the system. Understanding, visualizing, and learning large graphs is of key importance for understanding the underlying system and is a challenging task due to the data deluge problem. Our work here utilizes both spatial and temporal information from structured graphs. We learn spatial and temporal information using a specific type of neural network model. Our model is robust to the kind of graphs and their dynamics of evolution. Our approach is scalable to not only the size of the graph (number of vertices and edges) but also the number of attributes (features) of the data. We show that our approach is simple, generic, parallelizable, and performs at-par with the state-of-the-art techniques. We also compare the results of our model against other existing techniques. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.