Seminar Series | Combining Machine Learning with Traditional Optimization Strategies to Solve Large Scale Combinatorial Optimization Problems

Speaker: Katie McConky, Rochester Institute of Technology 

Title: Combining Machine Learning with Traditional Optimization Strategies to Solve Large Scale Combinatorial Optimization Problems

Abstract: Combinatorial optimization problems, while ubiquitous, are some of the most challenging problems to solve to optimality due to their exponentially growing decision space as problems sizes grow. In recent years, advances in artificial intelligence and machine learning have enabled solutions in other challenging spaces such as self driving vehicles, language modeling, and generative AI. In this talk we’ll explore several ways we have been integrating machine learning and traditional optimization strategies, such as metaheuristics and MILP solvers, to provide fast high quality solutions to multi-vehicle routing problems. We’ll first discuss GAMMA, a graph attention model for multi agent problems that uses deep reinforcement learning to learn complex heuristics for approximately solving team orienteering problems with non-Euclidean edges. Next we’ll discuss improvements that can be made to classic genetic algorithms by incorporating a learning mechanism to improve parent selection. Finally, we’ll discuss our Hybrid Learning Optimization Solver that combines machine learning, metaheuristics and traditional MILP solvers to collaboratively solve multi-agent problems.

Bio: Dr. Katie McConky is the Department Head of the Industrial and Systems Engineering department at the Rochester Institute of Technology. She holds a PhD in Industrial Engineering from University of Buffalo, an MS in Industrial and Systems Engineering from RIT, and a BS in Industrial Engineering from RIT. Her research passions lie at the intersection of operations research, data analytics and machine learning. Dr. Katie works in a broad range of application areas including energy related analytics and optimization, cyber attack prediction, vehicle routing under terrain uncertainty, and more recently with kidney exchange problems. Her work has been sponsored by numerous agencies including NYSERDA (New York State Energy Research and Development Authority), Office of Naval Research, NASA, AFRL, IARPA, and DARPA among others. She is currently a Technical Vice President of IISE, and the advisor of the RIT IISE Student Chapter. She enjoys teaching classes in operations research and forecasting, and has won department and college level teaching awards every year she has been eligible.