Abstract
The Optimal Transport (OT) problem investigates a transport map that connectstwo distributions while minimizing a given cost function. Finding such atransport map has diverse applications in machine learning, such as generativemodeling and image-to-image translation. In this paper, we introduce a scalableand simulation-free approach for solving the Entropic Unbalanced OptimalTransport (EUOT) problem. We derive the dynamical form of this EUOT problem,which is a generalization of the Schr\"odinger bridges (SB) problem. Based onthis, we derive dual formulation and optimality conditions of the EUOT problemfrom the stochastic optimal control interpretation. By leveraging theseproperties, we propose a simulation-free algorithm to solve EUOT, calledSimulation-free EUOT (SF-EUOT). While existing SB models require expensivesimulation costs during training and evaluation, our model achievessimulation-free training and one-step generation by utilizing the reciprocalproperty. Our model demonstrates significantly improved scalability ingenerative modeling and image-to-image translation tasks compared to previousSB methods.