Graph Neural Networks for Solving Combinatorial Optimization Problems

Mixed-integer linear programming problems (MILP) are solved for developing context-specific metabolic models in biotechnology and for determining operation strategies in water distribution networks in engineering. The MILP problems have to be solved periodically to find optimal solutions. However, large-scale MILP problems are difficult to solve and intractable in practice. Recently, Graph neural networks (GNN) based approaches have shown potential promises in exploiting solution patterns for solving these problems. This project will develop novel and efficient GNN-based approach for solving MILP problems. It will combine the complementary skills of Dr. David Tay in graph signal processing and of Dr. Nirav Bhatt and Prof. Sridharakumar Narasiman in solving MILPs for biology and infrastructure networks.