Variational Bayesian Optimal Experimental Design

TL;DR

This paper introduces fast variational Bayesian estimators for expected information gain in experimental design, significantly improving speed and accuracy, and demonstrating practical effectiveness through multiple experiments.

Contribution

It develops novel amortized variational inference-based estimators for EIG, enhancing efficiency and accuracy in Bayesian optimal experimental design.

Findings

Estimators outperform previous methods in speed and accuracy.

The approach is effective in real-world experimental scenarios.

Theoretical and empirical validation confirms improvements.

Abstract

Bayesian optimal experimental design (BOED) is a principled framework for making efficient use of limited experimental resources. Unfortunately, its applicability is hampered by the difficulty of obtaining accurate estimates of the expected information gain (EIG) of an experiment. To address this, we introduce several classes of fast EIG estimators by building on ideas from amortized variational inference. We show theoretically and empirically that these estimators can provide significant gains in speed and accuracy over previous approaches. We further demonstrate the practicality of our approach on a number of end-to-end experiments.