Graph Comparison using Deep Graph Representation Learning University of Leicester in United Kingdom

GTA funded PhD studentship in ComputingHighlightsDevelopment of novel deep graph-based methods for embedding graphs in a low-dimensional space.Combining classical graph embedding techniques with graph neural networks to achieve higher accuracy.Explore the applications of methods developed here to graph datasets obtained from diverse scientific domains.ProjectGraph-based methods have recently emerged as a powerful tool for analysing the structure of a complex system. Such systems involve data that lies on a non-Euclidean space and is not only describe by entities but also by the relationships between those entities. For instance, traditionally, individual cellular components and their functions are studied. But it is a well-known fact that most biological functions are due to interactions between different cellular constituents. This resulted in emergence of various biological networks such as protein-protein interaction networks, metabolic networks, and gene regulatory networks. However, since graphs are not vectors, one of the limitations with graph-based analysis is that the traditional machine learning techniques cannot be directly applied to graph-based data.The aim of this project is to develop novel deep graph-based methods that can be used to embed a graph in a Euclidean space, where standard machine learning techniques such as clustering and classification can be directly applied to graphs. Our earlier work was based on graph kernels that are one of the most widely used techniques for graph classification. However, in recent years, graph neural networks (GNN) are becoming more popular due to their superior performance. A GNN a special type of deep neural network specifically built to analyse graph-based data. It has been successfully applied to perform different tasks with graphs. For example, one of its variants, Deep Graph Convolutional Neural Network (DGCNN), combines a GNN with a Convolutional Neural Network (CNN) to perform graph classification and graph clustering. Our research will focus on combining the power of traditional graph kernels and GNN to develop more powerful ways of embedding graphs in a lower-dimensional feature space.Finally, we will be exploring the applications of the methods developed here in diverse scientific domains including computer vision, biomedical data, social network analysis, bioinformatics, and chemoinformatic. Enquiries to project supervisor   Dr. Furqan Aziz [email protected] or [email protected] details and application advice at https://le.ac.uk/study/research-degrees/funded-opportunities/cms-gta