Balancing Simulation-based Inference for Conservative Posteriors

TL;DR

This paper extends the balancing technique to any posterior density algorithm in simulation-based inference, aiming to produce more conservative and reliable posterior approximations across various methods.

Contribution

It introduces a balanced version of neural posterior estimation and contrastive neural ratio estimation, broadening the applicability of balancing in simulation-based inference.

Findings

Balanced algorithms produce more conservative posteriors

Empirical results show improved calibration across benchmarks

Alternative interpretation via $ ext{χ}^2$ divergence provided

Abstract

Conservative inference is a major concern in simulation-based inference. It has been shown that commonly used algorithms can produce overconfident posterior approximations. Balancing has empirically proven to be an effective way to mitigate this issue. However, its application remains limited to neural ratio estimation. In this work, we extend balancing to any algorithm that provides a posterior density. In particular, we introduce a balanced version of both neural posterior estimation and contrastive neural ratio estimation. We show empirically that the balanced versions tend to produce conservative posterior approximations on a wide variety of benchmarks. In addition, we provide an alternative interpretation of the balancing condition in terms of the $\chi^2$ divergence.