A1668
Title: Wasserstein Fisher Rao gradient flows: Sequential Monte Carlo and operator splitting
Authors: Sahani Pathiraja - University of New South Wales (Australia) [presenting]
Abstract: Wasserstein Fisher Rao (WFR) Gradient flows have recently been proposed for sampling from a target probability distribution, a task common across machine learning, Bayesian statistics, and statistical mechanics. Such Gradient flows have been shown to accelerate convergence over pure Wasserstein Gradient flows by combining the benefits of diffusive transport (Wasserstein) and reweighting (Fisher-Rao) type Gradient flows. WFR flows for the Kullback-Leibler divergence are connected to well-known Monte Carlo algorithms. This connection reveals a natural implementation using importance sampling and motivates a sequential Monte Carlo based algorithm for solving a WFR Gradient flow. Additionally, operator splitting can further accelerate convergence to equilibrium with a judiciously chosen order of operators.
