Supervision: Ali Hariri

Project type: Semester project (master) Master thesis

Available

Modern Graph Neural Networks (GNNs) have been used extensively in numerous applications, especially in Biology and Chemistry where data can be naturally represented as graphs. Despite their success, these GNN architectures come with numerous disadvantages such as scalability and expressiveness. The main target in this project is to focus on novel, out of the box techniques to advance graph representation learning methods.

Expected output: I will be working with the student together to investigate novel techniques to rewire graphs aiming to improve their learning on a set of tasks. The student will develop a solid understanding of message passing techniques and their current limitations, in addition to various creative ways to overcome them. Further information about the methodology and specific datasets to try will be discussed upon meeting the student.

The project can lead to a publication at a top AI venue.

Required Skills Experience with the PyTorch or JAX framework, familiarity with GNN, and a background or interest in computational biology would be beneficial.

Supervisor: Ali Hariri, a PhD student on Graph Neural Networks. For more information, please reach out to me at ali.hariri@epfl.ch.

References: A good overview as a starting point is the following paper:

Graph Neural Networks and Their Current Applications in Bioinformatics (Zhang et.al, 2021)