TOWARD EFFECTIVE NEURAL NETWORKS ON GRAPH DATA

Abstract

Graphs are ubiquitous in many practical settings, e.g., social networks, financial systems, transportation planning, online shopping recommendation, etc. Graph Neural Networks (GNNs) are a promising approach for learning from graph data, but they have yet to reach the successes of deep learning in natural language processing and computer vision. In this thesis, we explore to enhance the performance of neural networks for graph learning.

First, by using Data Augmentation (DA), we present a new method to enhance Graph Convolutional Networks (GCNs). Second, we propose the Mixup methods for two fundamental tasks in graph learning: node and graph classification. we present the MeTA (Memory Tower Augmentation) module: a multi-level module that processes the augmented graphs of different magnitudes on separate levels, and performs message passing across levels to provide adaptively augmented inputs for every prediction. Last but not least, we present a new neighbor sampling method on temporal graphs.