Sequential Neural Posterior and Likelihood Approximation

TL;DR

SNPLA is a simulation-based inference algorithm that uses normalizing flows to efficiently learn both likelihood and posterior distributions sequentially, avoiding MCMC and providing faster inference.

Contribution

SNPLA introduces a novel sequential normalizing flows-based method for joint likelihood and posterior approximation in implicit models, improving stability and speed.

Findings

Performs competitively with fewer simulations.

Generates posterior samples 4 orders of magnitude faster than MCMC.

Learns both likelihood and posterior jointly in a stable manner.

Abstract

We introduce the sequential neural posterior and likelihood approximation (SNPLA) algorithm. SNPLA is a normalizing flows-based algorithm for inference in implicit models, and therefore is a simulation-based inference method that only requires simulations from a generative model. SNPLA avoids Markov chain Monte Carlo sampling and correction-steps of the parameter proposal function that are introduced in similar methods, but that can be numerically unstable or restrictive. By utilizing the reverse KL divergence, SNPLA manages to learn both the likelihood and the posterior in a sequential manner. Over four experiments, we show that SNPLA performs competitively when utilizing the same number of model simulations as used in other methods, even though the inference problem for SNPLA is more complex due to the joint learning of posterior and likelihood function. Due to utilizing normalizing flows SNPLA generates posterior draws much faster (4 orders of magnitude) than MCMC-based methods.