Ensemble Kalman Inversion for General Likelihoods

TL;DR

This paper extends Ensemble Kalman inversion methods to a broader class of Bayesian models, allowing for likelihoods that can be sampled but not explicitly evaluated, thus broadening their applicability.

Contribution

It introduces a generalization of Ensemble Kalman inversion techniques to handle non-Gaussian likelihoods in Bayesian inference.

Findings

Ensemble Kalman inversion can be adapted to general likelihoods.

The method works even when likelihood densities are intractable.

The approach broadens the applicability of Ensemble Kalman methods.

Abstract

In this letter we generalise Ensemble Kalman inversion techniques to general Bayesian models where previously they were restricted to additive Gaussian likelihoods - all in the difficult setting where the likelihood can be sampled from, but its density not necessarily evaluated.