Learning to Bound Using Decision Diagrams and Reinforcement Learning

Explore a cutting-edge approach to finding tight bounds for discrete optimization problems in this DS4DM Coffee Talk. Delve into the world of decision diagrams and their role in obtaining superior upper and lower bounds compared to traditional methods like linear relaxations. Discover how the variable ordering significantly impacts the quality of bounds achieved through decision diagrams, and learn about the challenges in optimizing this ordering. Gain insights into a novel generic approach that leverages deep reinforcement learning to determine an optimal ordering for tightening bounds obtained with approximate decision diagrams. Understand how these improved bounds can be effectively utilized to enhance the performance of branch-and-bound algorithms, potentially revolutionizing solution methods for complex optimization problems.

Learning to Bound Using Decision Diagrams and Reinforcement Learning, Quentin Cappart