Discrete Optimization Meets Machine Learning
Discrete Optimization algorithms underlie intelligent decision-making in a wide variety of domains. From airline fleet scheduling to kidney exchanges and data center resource management, decisions are often modeled with binary on/off variables that are subject to operational and financial constraints.
In this talk, I introduce “Data-Driven Algorithm Design”, a novel paradigm for boosting the performance of discrete optimization algorithms by leveraging two types of data: the set of problem instances arising from the application of interest; and information generated while solving each instance. I will present Machine Learning (ML) approaches that have advanced the state-of-the-art in both exact integer programming solvers as well as heuristic algorithms, with a focus on applications arising in Computational Sustainability.
I will also share my vision towards establishing ML as a central component of the algorithm design process, one that complements human ingenuity rather than replace it. This poses a variety of theoretical, modeling and practical research questions in ML. Conversely, infusing constrained reasoning into ML can help address a number of pressing challenges relating to the robustness, fairness and interpretability of ML models; I will touch on my recent work in this space and a path towards closing the loop between Discrete Optimization and ML.
Elias Khalil is a Ph.D. candidate at the College of Computing at Georgia Tech. His research interests are in Artificial Intelligence with a focus on machine learning and discrete optimization. He is the recipient of an IBM Ph.D. Fellowship (2016-2017), the First Prize in the poster competition at INFORMS (2017) and the Best Paper Award at the NIPS Workshop on Frontiers of Network Analysis (2013). He has interned at IBM Research and Symantec Research Labs, and has received his MS from Georgia Tech (2014) and his BS from the American University of Beirut (2012), both in Computer Science.