The UTC Graduate School is pleased to announce that Elvis Frimpong will present Master’s research titled, A Learning-Enhanced Location Covering Approach to Endangered Species Conservation on 10/07/2026 at 3:30PM in Lupton Hall, Room 302. Everyone is invited to attend.
Mathematics
Chair: Dr. Lakmali Weerasena
Co-Chair:
Abstract:
Selecting habitat reserves or monitoring stations for endangered species requires identifying a minimum-cost set of locations that covers high-risk areas and maintains spatial connectivity. Standard covering formulations often produce fragmented solutions that do not support effective wildlife movement. We address this challenge with a hybrid learning-and-optimization framework that combines graph neural networks with exact optimization. By representing coverage and spatial connectivity within a unified graph, our framework trains the neural network via a self-supervised consensus of cost-connectivity-aware heuristics, eliminating the need for computationally expensive optimal training data. At inference, the learned model generates a connected feasible solution that provides a warm start for an exact solver, accelerating convergence while preserving global optimality. We evaluate the framework on synthetic landscapes and a real-world conservation planning application. Results show substantial reductions in the exact solution time compared to cold starts and standard heuristics, as well as a robust transfer from synthetic training instances to unseen conservation scenarios.