Data‐Driven Optimal Transport

Abstract

The problem of optimal transport between two distributions ρ(x) and μ(y) is extended to situations where the distributions are only known through a finite number of samples {xi} and {yj}. A weak formulation is proposed, based on the dual of the Kantorovich formulation, with two main modifications: replacing the expected values in the objective function by their empirical means over the {xi} and {yj}, and restricting the dual variables u(x) and v(y) to a suitable set of test functions adapted to the local availability of sample points. A procedure is proposed and tested for the numerical solution of this problem, based on a fluidlike flow in phase space, where the sample points play the role of active Lagrangian markers. © 2016 Wiley Periodicals, Inc.

Rejection-associated Mitochondrial Impairment After Heart Transplantation

“…In addition to the ISHLT, we used a 3-class system we developed following an unsupervised analysis described elsewhere. 23 - 25 Briefly, the output that the procedure provides is the matrix P0 (class, sample) containing the probability that each sample belongs to each class, and the subset of genes whose expression is most consistent with the classification found. To assign samples to classes, we evolve from a noninformative small random perturbation of the uniform assignment to its final value through a Bayesian procedure that uses the expression of each gene as new evidence to relax P0, thought of as a prior, toward the corresponding posterior.…”

Explanation of Variability and Removal of Confounding Factors from Data through Optimal Transport

“…The dual formulation in (3.8) involves the distributions k .x/ only through integrals representing the expected values of the k .x/ under k (a fact exploited in [24] to implement regular data-driven optimal transport). Thus, if only samples from k are available, it is natural to replace these expected values by empirical means:…”

“…For instance, one could extend the flow-based procedure developed for the data-driven optimal transport problem [24]. For the remainder of this article, we will instead adopt two much simpler, nonadaptive poorman procedures.…”

“…The data-based barycenter problem can be solved with a level of accuracy that adjusts to the number of samples available by defining the space F of allowed test functions adaptably: in areas where there are more sample points, the functions in F should have finer bandwidths. For instance, one could extend the flow-based procedure developed for the data-driven optimal transport problem [24]. For the remainder of this article, we will instead adopt two much simpler, nonadaptive poorman procedures.…”

Sample‐Based Optimal Transport and Barycenter Problems

“…Data-driven formulations take as input not the distributions 1;2 but sample sets from both. Methodologies proposed include a fluid-flow-like algorithm [44] and an adaptive linear programming approach [9].…”