Pairwise Multi-marginal Optimal Transport via Universal Poisson Coupling

by   Cheuk Ting Li, et al.

We investigate the problem of pairwise multi-marginal optimal transport, that is, given a collection of probability distributions P_1,...,P_m on a Polish space, to find a coupling X_1,...,X_m, X_i∼ P_i, such that E[c(X_i,X_j)]< r_X∼ P_i, Y∼ P_jE[c(X,Y)] for any i,j, where c is a cost function and r>1. In other words, every pair (X_i,X_j) has an expected cost at most a factor of r from its lowest possible value. This can be regarded as a multi-agent matching problem with a fairness requirement. For the cost function c(x,y)= x-y_2^q over R^n, where q>0, we show that a finite r is attainable when either n=1 or q<1, and not attainable when n>2 and q>1. It is unknown whether such r exists when n>2 and q=1. Also, we show that r grows at least as fast as Ω(n^1{q=1}+q/2) when n→∞. The case of the discrete metric cost c(x,y)=1{x≠ y}, and more general metric and ultrametric costs are also investigated.



page 1

page 2

page 3

page 4


Multi-marginal optimal transport and probabilistic graphical models

We study multi-marginal optimal transport problems from a probabilistic ...

𝒲_∞-transport with discrete target as a combinatorial matching problem

In this short note, we show that given a cost function c, any coupling π...

Metric properties of homogeneous and spatially inhomogeneous F-divergences

In this paper I investigate the construction and the properties of the s...

Efficient robust optimal transport: formulations and algorithms

The problem of robust optimal transport (OT) aims at recovering the best...

Distributionally Robust Martingale Optimal Transport

We study the problem of bounding path-dependent expectations (within any...

Factored couplings in multi-marginal optimal transport via difference of convex programming

Optimal transport (OT) theory underlies many emerging machine learning (...

Probabilistic Inverse Optimal Transport

Optimal transport (OT) formalizes the problem of finding an optimal coup...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.