We present an innovative Wasserstein Gradient Flow algorithm for solving optimization problems over the space of probability densities. Differently form other state of the art methods, the proposed approach does not involve any computationally expensive training of neural networks. The main contribution is a novel parametrization of the transport map that allows to recast the widely adopted JKO schema from a scalarized bi-objective to a constrained optimization problem. This provides three relevant advantages: a better control of the desired quality of approximation of the optimal transport map; convergence to the optimum without affecting the quality of the transport map; and a significantly lower computational cost (i.e., runtime-per-iteration). Experimental results on a set of well diversified test cases provide the empirical evidence of the advantages offered by the proposed approach.

Candelieri, A., Ponti, A., Archetti, F. (2025). A Constrained-JKO Scheme for Effective and Efficient Wasserstein Gradient Flows. In Learning and Intelligent Optimization 18th International Conference, LION 18, Ischia Island, Italy, June 9–13, 2024, Revised Selected Papers (pp.66-80). Springer Science and Business Media Deutschland GmbH [10.1007/978-3-031-75623-8_6].

A Constrained-JKO Scheme for Effective and Efficient Wasserstein Gradient Flows

Candelieri A.;Ponti A.;Archetti F.
2025

Abstract

We present an innovative Wasserstein Gradient Flow algorithm for solving optimization problems over the space of probability densities. Differently form other state of the art methods, the proposed approach does not involve any computationally expensive training of neural networks. The main contribution is a novel parametrization of the transport map that allows to recast the widely adopted JKO schema from a scalarized bi-objective to a constrained optimization problem. This provides three relevant advantages: a better control of the desired quality of approximation of the optimal transport map; convergence to the optimum without affecting the quality of the transport map; and a significantly lower computational cost (i.e., runtime-per-iteration). Experimental results on a set of well diversified test cases provide the empirical evidence of the advantages offered by the proposed approach.
paper
JKO; Optimal Transport; Wasserstein Gradient Flow;
English
18th International Conference, LION 18 - June 9–13, 2024
2024
Festa, P; Ferone, D; Pastore, T; Pisacane, O
Learning and Intelligent Optimization 18th International Conference, LION 18, Ischia Island, Italy, June 9–13, 2024, Revised Selected Papers
9783031756221
2025
14990
66
80
none
Candelieri, A., Ponti, A., Archetti, F. (2025). A Constrained-JKO Scheme for Effective and Efficient Wasserstein Gradient Flows. In Learning and Intelligent Optimization 18th International Conference, LION 18, Ischia Island, Italy, June 9–13, 2024, Revised Selected Papers (pp.66-80). Springer Science and Business Media Deutschland GmbH [10.1007/978-3-031-75623-8_6].
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/10281/551729
Citazioni
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
Social impact