Skip to main content

Robust and Interpretable Deep Graph Learning

Primary supervisor

Shirui Pan

Graph Neural networks (GNNs) or Deep Graph Learning are new techniques which enable deep learning to perform on graph or structure data. Due to its superb ability in many applications, including social networks, communication networks, and knowledge graphs, GNNs have attracted increasing attention in the research community. However, existing GNN studies suffer from two limitations: 1) they are vulnerable to the external attack. A simple modification of the graph structure and or features will make the GNN models fail. 2) like most deep learning models, they are typically black-box models and are not interpretable.

This project aims to solve these two challenges. A number of robust and interpretable deep graph learning methods will be developed and applied to knowledge graphs, traffic networks, and social networks. etc.

Required knowledge

  • Machine Learning Backgrounds
  • Deep Learning Experience
  • Familiar with Python, Pytorch, Tensorflow

Project funding


Learn more about minimum entry requirements.