Daniel Mimouni | Searcharxiv

arXiv2024

Scenario Tree Reduction via Wasserstein Barycenters

Scenario tree reduction techniques are essential for achieving a balance between an accurate representation of uncertainties and computational complexity when solving multistage stochastic programming problems. In the realm of available techniques, the Kovacevic and Pichler algorithm (Ann. Oper. Res., 2015 [1]) stands out for employing the nested distance, a metric for comparing multistage scenario trees. However, dealing with large-scale scenario trees can lead to a prohibitive computational burden due to the algorithm's requirement of solving several large-scale linear problems per iteration. This study concentrates on efficient approaches to solving such linear problems, recognizing that their solutions are Wasserstein barycenters of the tree nodes' probabilities on a given stage. We leverage advanced optimal transport techniques to compute Wasserstein barycenters and significantly improve the computational performance of the Kovacevic and Pichler algorithm. Our boosted variants of this algorithm are benchmarked on several multistage scenario trees. Our experiments show that compared to the original scenario tree reduction algorithm, our variants can be eight times faster for reducing scenario trees with 8 stages, 78 125 scenarios, and 97 656 nodes.

math.OC↗

arXiv2023

Computing Wasserstein Barycenter via operator splitting: the method of averaged marginals

The Wasserstein barycenter (WB) is an important tool for summarizing sets of probability measures. It finds applications in applied probability, clustering, image processing, etc. When the measures' supports are finite, computing a (balanced) WB can be done by solving a linear optimization problem whose dimensions generally exceed standard solvers' capabilities. In the more general setting where measures have different total masses, we propose a convex nonsmooth optimization formulation for the so-called unbalanced WB problem. Due to their colossal dimensions, we introduce a decomposition scheme based on the Douglas-Rachford splitting method that can be applied to both balanced and unbalanced WB problem variants.Our algorithm, which has the interesting interpretation of being built upon averaging marginals, operates a series of simple (and exact) projections that can be parallelized and even randomized, making it suitable for large-scale datasets. Numerical comparisons against state-of-the-art methods on several data sets from the literature illustrate the method's performance.

math.OC↗