Ezra's Round Table / Systems Seminar: Feng Qiu (Argonne National Laboratory)
Frank H. T. Rhodes Hall 253
A Learning-Enhanced Optimization Framework for Routinely Solved Optimization Problems
Most real-world applications solve optimization problems routinely. These routinely solved optimization problems share significant similarity in problem structure and input data. In this talk, we will introduce a machine learning framework that explores the similarity among the routinely solved optimization problems (specifically, mixed-integer programming) and help the solvers solve the MIPs faster and faster as more instances are solved. We will use security constrained unit commitment as an example to demonstrate this learning-enhanced optimization framework and show effectiveness on a set of challenging combinatorial optimization problems. Lastly, we will introduce an open-source package that implements the learning-enhanced optimization framework: MIPLearn.
Dr. Feng Qiu received his Ph.D. from the School of Industrial and Systems Engineering at the Georgia Institute of Technology in 2013. He is a principal computational scientist and a section leader with the Energy Systems Division at Argonne National Laboratory. His current research interests include power system modeling and optimization, electricity markets, power grid resilience, machine learning and data analytics.