Adaptive Partitioning for Chance-Constrained Problems - DS4DM Coffee Talk

Explore an innovative approach to solving chance-constrained stochastic optimization problems (CCSPs) with finite support in this 22-minute DS4DM Coffee Talk presented by Marius Roland from Polytechnique Montréal, Canada. Delve into an iterative algorithm that tackles reduced-size chance-constrained models through scenario set partitioning, yielding bounds on the optimal objective value of the original CCSP. Discover the key operations of refinement and merging that drive the algorithm's efficiency, and learn how these processes improve bounds while minimizing model size increases. Examine the theoretical foundations that guarantee strict bound improvements and finite termination at an optimal solution. Gain insights into computational enhancements, partition initialization strategies, and connections to quantile cuts that lead to stronger valid inequalities. Evaluate the algorithm's performance through numerical experiments on chance-constrained multidimensional knapsack problems, comparing it to state-of-the-art methods and analyzing the impact of each component.

Adaptive Partitioning for Chance-Constrained Problems, Marius Roland