Post-hoc loss-calibration for Bayesian neural networks
Meet P. Vadera, Soumya Ghosh, Kenney Ng, Benjamin M. Marlin
Abstract
Bayesian decision theory provides an elegant framework for acting optimally under uncertainty when tractable posterior distributions are available. Modern Bayesian models, however, typically involve intractable posteriors that are approximated with, potentially crude, surrogates. This difficulty has engendered loss-calibrated techniques that aim to learn posterior approximations that favor high-utility decisions. In this paper, focusing on Bayesian neural networks, we develop methods for correcting approximate posterior predictive distributions encouraging them to prefer high-utility decisions. In contrast to previous work, our approach is agnostic to the choice of the approximate inference algorithm, allows for efficient test time decision making through amortization, and empirically produces higher quality decisions. We demonstrate the effectiveness of our approach through controlled experiments spanning a diversity of tasks and datasets.
BibTeX
@InProceedings{pmlr-v161-vadera21a,
title = {Post-hoc loss-calibration for Bayesian neural networks},
author = {Vadera, Meet P. and Ghosh, Soumya and Ng, Kenney and Marlin, Benjamin M.},
booktitle = {Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence},
pages = {1403--1412},
year = {2021},
editor = {de Campos, Cassio and Maathuis, Marloes H.},
volume = {161},
series = {Proceedings of Machine Learning Research},
month = {27--30 Jul},
publisher = {PMLR},
pdf = {https://proceedings.mlr.press/v161/vadera21a/vadera21a.pdf},
url = {https://proceedings.mlr.press/v161/vadera21a.html},
abstract = {Bayesian decision theory provides an elegant framework for acting optimally under uncertainty when tractable posterior distributions are available. Modern Bayesian models, however, typically involve intractable posteriors that are approximated with, potentially crude, surrogates. This difficulty has engendered loss-calibrated techniques that aim to learn posterior approximations that favor high-utility decisions. In this paper, focusing on Bayesian neural networks, we develop methods for correcting approximate posterior predictive distributions encouraging them to prefer high-utility decisions. In contrast to previous work, our approach is agnostic to the choice of the approximate inference algorithm, allows for efficient test time decision making through amortization, and empirically produces higher quality decisions. We demonstrate the effectiveness of our approach through controlled experiments spanning a diversity of tasks and datasets.}
}