Sequential Neural Methods for Likelihood-free Inference

TL;DR

This paper compares neural methods for likelihood-free inference, focusing on how they select simulations and learn either approximate posteriors or surrogate likelihoods, to identify the most effective strategies.

Contribution

It provides a direct comparison of neural approaches for likelihood-free inference, clarifying the impact of different simulation selection and learning strategies.

Findings

Neural methods can outperform traditional ABC with fewer simulations

Choosing between posterior and surrogate likelihood learning affects performance

Controlled comparisons highlight best practices for neural likelihood-free inference

Abstract

Likelihood-free inference refers to inference when a likelihood function cannot be explicitly evaluated, which is often the case for models based on simulators. Most of the literature is based on sample-based `Approximate Bayesian Computation' methods, but recent work suggests that approaches based on deep neural conditional density estimators can obtain state-of-the-art results with fewer simulations. The neural approaches vary in how they choose which simulations to run and what they learn: an approximate posterior or a surrogate likelihood. This work provides some direct controlled comparisons between these choices.