FLAIR · project
Multimarginal Optimal Transport with a Tree-Structured Cost and the Schrödinger Bridge Problem
Information fusion and tracking often require coupling several probability distributions at once. Multimarginal optimal transport provides a framework for this task, but a direct representation of the joint transport plan can become prohibitively large. This work exploits costs that decompose along the edges of a tree.
Using the graph structure
Entropy regularization connects the problem to a Schrödinger bridge on the same tree. The resulting structure allows the computational steps to be expressed through matrix–vector multiplications of relatively small dimensions, avoiding direct operations on a full high-dimensional transport tensor.
Estimation with less artificial diffusion
The paper compares joint multimarginal regularization with regularizing pairwise transport problems separately. Joint regularization introduces less diffusion in the studied settings, preserving sharper distributional information. Numerical examples and tracking of an ensemble of indistinguishable agents illustrate how the framework can recover population-level evolution from distributional observations.
Publication
Multimarginal Optimal Transport with a Tree-Structured Cost and the Schrödinger Bridge Problem
Isabel Haasler, Axel Ringh, Yongxin Chen, and Johan Karlsson
SIAM Journal on Control and Optimization, 59(4), 2428–2453, 2021.