Correcting Predictions for Approximate Bayesian Inference
Abstract
Bayesian models quantify uncertainty and facilitate optimal decision-making in downstream applications. For most models, however, practitioners are forced to use approximate inference techniques that lead to sub-optimal decisions due to incorrect posterior predictive distributions. We present a novel approach that corrects for inaccuracies in posterior inference by altering the decision-making process. We train a separate model to make optimal decisions under the approximate posterior, combining interpretable Bayesian modeling with optimization of direct predictive accuracy in a principled fashion. The solution is generally applicable as a plug-in module for predictive decision-making for arbitrary probabilistic programs, irrespective of the posterior inference strategy. We demonstrate the approach empirically in several problems, confirming its potential.
Cite
@article{arxiv.1909.04919,
title = {Correcting Predictions for Approximate Bayesian Inference},
author = {Tomasz Kuśmierczyk and Joseph Sakaya and Arto Klami},
journal= {arXiv preprint arXiv:1909.04919},
year = {2019}
}