Abstract:In this paper, we study efficient approximate sampling for probability distributions known up to normalization constants. We specifically focus on a problem class arising in Bayesian inference for large-scale inverse problems in science and engineering applications. The computational challenges we address with the proposed methodology are: (i) the need for repeated evaluations of expensive forward models; (ii) the potential existence of multiple modes; and (iii) the fact that gradient of, or adjoint solver for, the forward model might not be feasible. While existing Bayesian inference methods meet some of these challenges individually, we propose a framework that tackles all three systematically. Our approach builds upon the Fisher-Rao gradient flow in probability space, yielding a dynamical system for probability densities that converges towards the target distribution at a uniform exponential rate. This rapid convergence is advantageous for the computational burden outlined in (i). We apply Gaussian mixture approximations with operator splitting techniques to simulate the flow numerically; the resulting approximation can capture multiple modes thus addressing (ii). Furthermore, we employ the Kalman methodology to facilitate a derivative-free update of these Gaussian components and their respective weights, addressing the issue in (iii). The proposed methodology results in an efficient derivative-free posterior approximation method, flexible enough to handle multi-modal distributions: Gaussian Mixture Kalman Inversion (GMKI). The effectiveness of GMKI is demonstrated both theoretically and numerically in several experiments with multimodal target distributions, including proof-of-concept and two-dimensional examples, as well as a large-scale application: recovering the Navier-Stokes initial condition from solution data at positive times.

Understanding posterior projection effects with normalizing flows

Normalizing Flows for Bayesian Posteriors: Reproducibility and Deployment

Latent Variable Modelling with Hyperbolic Normalizing Flows

Marginal Post Processing of Bayesian Inference Products with Normalizing Flows and Kernel Density Estimators

$\mathtt{emuflow}$: Normalising Flows for Joint Cosmological Analysis

Learned harmonic mean estimation of the marginal likelihood with normalizing flows

Learned harmonic mean estimation of the Bayesian evidence with normalizing flows

Fast Posterior Probability Sampling with Normalizing Flows and Its Applicability in Bayesian analysis in Particle Physics

emuflow: normalizing flows for joint cosmological analysis

Bernstein flows for flexible posteriors in variational Bayes

Efficient, multimodal, and derivative-free Bayesian inference with Fisher-Rao gradient flows

Flow Matching for Scalable Simulation-Based Inference

A Tale of Two Latent Flows: Learning Latent Space Normalizing Flow with Short-run Langevin Flow for Approximate Inference

Injective flows for star-like manifolds

Normalising Flows for Bayesian Gravity Inversion

Learning to solve Bayesian inverse problems: An amortized variational inference approach using Gaussian and Flow guides

Stable Training of Normalizing Flows for High-dimensional Variational Inference

Comparative Study of Coupling and Autoregressive Flows through Robust Statistical Tests