Neural Approximate Sufficient Statistics for Implicit Models

TL;DR

This paper introduces a neural network-based method to automatically learn sufficient statistics for implicit generative models, enhancing the performance of likelihood-free inference techniques without requiring density estimation.

Contribution

It proposes an infomax-based approach to learn mutual information maximizing representations, improving summary statistic construction for implicit models.

Findings

Boosts performance of ABC and neural likelihood methods

Does not require density or density ratio estimation

Applicable to various inference tasks

Abstract

We consider the fundamental problem of how to automatically construct summary statistics for implicit generative models where the evaluation of the likelihood function is intractable, but sampling data from the model is possible. The idea is to frame the task of constructing sufficient statistics as learning mutual information maximizing representations of the data with the help of deep neural networks. The infomax learning procedure does not need to estimate any density or density ratio. We apply our approach to both traditional approximate Bayesian computation and recent neural likelihood methods, boosting their performance on a range of tasks.