This paper presents primal heuristics for the computation of Wasserstein Barycenters of a given set of discrete probability measures. The computation of a Wasserstein Barycenter is formulated as an optimization problem over the space of discrete probability measures. In practice, the barycenter is a discrete probability measure which minimizes the sum of the pairwise Wasserstein distances between the barycenter itself and each input measure. While this problem can be formulated using Linear Programming techniques, it remains a challenging problem due to the size of real-life instances. In this paper, we propose simple but efficient primal heuristics, which exploit the properties of the optimal plan obtained while computing the Wasserstein Distance between a pair of probability measures. In order to evaluate the proposed primal heuristics, we have performed extensive computational tests using random Gaussian distributions, the MNIST handwritten digit dataset, and the Fashion MNIST dataset introduced by Zalando. We also used Translated MNIST, a modification of MNIST which contains original images, rescaled randomly and translated into a larger image. We compare the barycenters computed by our heuristics with the exact solutions obtained with a commercial Linear Programming solver, and with a state-of-the-art algorithm based on Gaussian convolutions. Our results show that the proposed heuristics yield in very short run time and an average optimality gap significantly smaller than 1%.

Primal Heuristics for Wasserstein Barycenters

Bouchet, Pierre-Yves
Writing – Original Draft Preparation
;
Gualandi, Stefano
Writing – Original Draft Preparation
;
2020-01-01

Abstract

This paper presents primal heuristics for the computation of Wasserstein Barycenters of a given set of discrete probability measures. The computation of a Wasserstein Barycenter is formulated as an optimization problem over the space of discrete probability measures. In practice, the barycenter is a discrete probability measure which minimizes the sum of the pairwise Wasserstein distances between the barycenter itself and each input measure. While this problem can be formulated using Linear Programming techniques, it remains a challenging problem due to the size of real-life instances. In this paper, we propose simple but efficient primal heuristics, which exploit the properties of the optimal plan obtained while computing the Wasserstein Distance between a pair of probability measures. In order to evaluate the proposed primal heuristics, we have performed extensive computational tests using random Gaussian distributions, the MNIST handwritten digit dataset, and the Fashion MNIST dataset introduced by Zalando. We also used Translated MNIST, a modification of MNIST which contains original images, rescaled randomly and translated into a larger image. We compare the barycenters computed by our heuristics with the exact solutions obtained with a commercial Linear Programming solver, and with a state-of-the-art algorithm based on Gaussian convolutions. Our results show that the proposed heuristics yield in very short run time and an average optimality gap significantly smaller than 1%.
2020
Integration of Constraint Programming, Artificial Intelligence, and Operations Research
Hebrard E., Musliu N.
Engineering Mathematics covers resources on applied mathematics, mathematical modelling, combinatorics, optimization techniques, numerical methods, and statistical methods that have an emphasis on engineering systems.
Esperti anonimi
Inglese
contributo
International conference on Integration of Constraint Programming, Artificial Intelligence, and Operations Research
Mon 21 Sep 2020 - Thu 24 Sep 2020
Vienna
Internazionale
STAMPA
12296
239
255
17
978-3-030-58941-7
978-3-030-58942-4
Springer
none
Bouchet, Pierre-Yves; Gualandi, Stefano; Rousseau, Louis-Martin
273
info:eu-repo/semantics/conferenceObject
3
4 Contributo in Atti di Convegno (Proceeding)::4.1 Contributo in Atti di convegno
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11571/1348920
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 4
  • ???jsp.display-item.citation.isi??? ND
social impact