A tour of Lagrangian duality

Session date: 16 December 2024

Session host: Konstantin Sidorov

Summary:

In this session, I plan to follow up on our recent discussions and do a deeper dive into the concept of Lagrangian duality, a technique for relaxing an optimization problem by moving the constraints into the objective and addressing the resulting min-max problem. As this technique has a wide range of applications, I intend to show them in a sequence ranging from “canonical” convex optimization ideas to modern applications that progress towards a generic, learnable model for bounding in constraint programming. I hope that by the end of this session, you will have a robust view of both the technique and its success stories!