Grant: Discovery Grants Program - Individual

Note: this research does not involve Indigenous communities, societies, or individuals. Summary: Multiplex graphs, in which nodes are connected by multiple edges, provide rich and flexible modeling of complex interaction scenarios among entities of real-world systems. With nodes representing entities and edges reflecting their interactions, many natural and engineered systems can be described using multiplex graphs. For example, protein-protein interactions can be captured by various protocols. Different protocols lead to multiple interactions, resulting in a multiplex graph with many types of connections between two proteins (nodes). These different types of interactions are called dimensions of the multiplex graph. Valuable information on nodes' interactions can be hidden across various combinations of dimensions. The high dimensionality, the large size, the dynamic nature of multiplex graphs, and the heterophilic, as well as the homophilic patterns they encompass, make them notoriously difficult to mine, which in turn pose new challenges for the elaboration of effective learning algorithms. My program deals with the scarcity of algorithmic methods for mining such complex graphs and will handle original and challenging problems that give rise to novel multiplex graph analysis paradigms. In these settings, my long-term goal is to elaborate graph neural network theories and models to fully harness the potential of heterophilic large-scale high-dimensional multiplex graphs. In summary, my short-term objectives consist of (1) Solving the highly curved manifold problem, (2) Addressing scalability challenges, (3) Tackling the heterophily phenomenon, and (4) Handling dynamicity. By achieving these objectives, my research will advance the state-of-the-art by creating knowledge and insight that provide a basis for action to enhance multiplex graph representation learning for various graph mining tasks. The originality of my program resides in tackling challenges that have not yet been addressed in the context of high-dimensional multiplex graphs. In this context, our contributions will have significant theoretical and practical impacts. On the theoretical side, my students and I will devise effective solutions through an in-depth investigation of complex and fundamental problems (manifold geometry, scalability, heterophily, and dynamicity) that have been overlooked by prior work in the field. My research will thus bring new perspectives on existing research on graph neural networks and open up various new research opportunities. On the practical side, we will develop open-source tools and methods to learn general-purpose representations for multiplex graphs efficiently. I target reproducible research results and practical solutions to concrete real-world problems. The proposed research will train 3 PhD, 4 MSc, and 5 BSc who will gain expertise in artificial intelligence that meets the needs of academia, Canadian business, and public sectors.