Speaker: Yingfeng Wang, University of Tennessee at Chattanooga

Title: Using Noise Injection to Improve Graph Autoencoder Training

Abstract: Autoencoder has been widely applied in many areas such as dimension reduction, image processing, and natural language processing. Recently, researchers have developed graph autoencoders to process graph data. The training of graph autoencoders aims to minimize the difference between the input graphs and reconstructed graphs. This presentation will introduce a noise injection strategy for improving graph autoencoder training. This strategy adds perturbations to the input data to overcome overfitting. Since only training data are changed, this strategy can fit most training algorithms.