Frontiers in Graph Representation Learning

Supervisor

Suitable for

Abstract

There are various projects available in the theme of ‘Graph Representation Learning’ (e.g., graph neural networks), please get in touch for the most up-to-date information. There are multiple projects aiming for (i) designing new graph neural network models to overcome some well-known existing limitations of graph neural networks (e.g., oversmoothing, oversquashing, inexpressiveness), (ii) improving our overall theoretical understanding of graph representation learning models (foundational aspects related to probability theory, graph theory, logic, as well as algorithmic aspects), (iii) improving the inductive bias of graph neural networks for machine learning over different types of graphs (e.g., directed, relational graphs), (iv) exploring tasks beyond widely studied ones, such as graph generation, from a foundational aspect (v) novel applications of graph represenation learning. It is important to identify a project which matches the students background and so the details are somewhat subtle to be included in a project description. There are various industrial and academic collaborators, depending on the specific direction. Most of the necessary background information is covered in the graph representation learning course offered in the Department in MT.