"On the use of Machine Learning in Discrete Optimization"
Andrea Lodi

In this talk, I will review the use of Machine Learning (ML) in Discrete Optimization (DO). The talks covers a wide range of methods. At the extremes, on the one side, I will discuss the use of ML directly as a way of generating heuristic solutions for DO problems for which I will try to highlight the connections among known exact and heuristic paradigms like dynamic programming, reinforcement learning and agent-based metaheuristics. On the other side, I will discuss the more capillary use of ML techniques to replace DO algorithmic blocks that are either too time consuming or would benefit from a stronger statistical viewpoint.