Automatic Posterior Transformation for Likelihood-Free Inference

TL;DR

The paper introduces Automatic Posterior Transformation (APT), a flexible and scalable neural inference method that improves Bayesian inference for complex simulators with intractable likelihoods, handling high-dimensional data effectively.

Contribution

It presents APT, a novel sequential neural posterior estimation technique that adapts to arbitrary proposals and integrates flow-based density estimators for enhanced likelihood-free inference.

Findings

APT outperforms existing methods in flexibility and scalability.

It effectively handles high-dimensional data like time series and images.

APT demonstrates improved inference accuracy in complex simulation scenarios.

Abstract

How can one perform Bayesian inference on stochastic simulators with intractable likelihoods? A recent approach is to learn the posterior from adaptively proposed simulations using neural network-based conditional density estimators. However, existing methods are limited to a narrow range of proposal distributions or require importance weighting that can limit performance in practice. Here we present automatic posterior transformation (APT), a new sequential neural posterior estimation method for simulation-based inference. APT can modify the posterior estimate using arbitrary, dynamically updated proposals, and is compatible with powerful flow-based density estimators. It is more flexible, scalable and efficient than previous simulation-based inference techniques. APT can operate directly on high-dimensional time series and image data, opening up new applications for likelihood-free inference.