Skip to Main content Skip to Navigation
Preprints, Working Papers, ...

Stability of Semi-Dual Unbalanced Optimal Transport: fast statistical rates and convergent algorithm.

Abstract : In this paper, we derive stability results for the semi-dual formulation of unbalanced optimal transport. From a statistical point of view, the gain of stability with respect to the balanced case allows to employ localization arguments while only assuming strong convexity of potentials and recover superparametric rates. Then we derive a provably convergent theoretical algorithm to minimize the semi-dual: if the potentials are constrained to be strongly convex, both the values and minimizers converge at a 1/k rate. Under an additional smoothness assumption, the convergence is exponential in the balanced case. Finally we instantiate a tractable version of our theoretical algorithm in the case of strongly convex, possibly smooth potentials. We benchmark the method in the balanced case on a 2D experiment and in the unbalanced case on a medium dimension synthetic experiment.
Document type :
Preprints, Working Papers, ...
Complete list of metadata


https://hal.archives-ouvertes.fr/hal-03609629
Contributor : François-Xavier Vialard Connect in order to contact the contributor
Submitted on : Wednesday, June 15, 2022 - 10:43:30 PM
Last modification on : Friday, June 24, 2022 - 4:05:31 AM

Files

main3.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-03609629, version 2
  • ARXIV : 2203.09143

Citation

Adrien Vacher, François-Xavier Vialard. Stability of Semi-Dual Unbalanced Optimal Transport: fast statistical rates and convergent algorithm.. 2022. ⟨hal-03609629v2⟩

Share

Metrics

Record views

42

Files downloads

16