MP1 - Plénière 1 / Plenary 1

In many industry settings, including the electrical power grid, supply chains, manufacturing, and transportation networks, the same optimization problem is solved repeatedly for instances taken from a distribution that can be learned or forecasted. The scale and complexity of these applications have grown significantly in recent years, challenging traditional optimization approaches. This talk studies how to accelerate the solving of these parametric optimization problems to meet real-time constraints present in many applications. It first reviews the concept of optimization proxies that learn the input/output mappings of parametric optimization problems, computing near-optimal feasible solutions and providing quality guarantees. The talk also presents how to "learn to optimize" highly complex optimization problems, fusing optimization methodologies with supervised learning and reinforcement learning. The methodologies are highlighted on industrial problems in grid optimization, end-to-end supply chains, logistics, and transportation systems. They reveal beautiful connections between machine learning and optimization, leveraging fundamental theoretical results to push the practice of optimization.