Traditional training of deep classifiers yields overconfident models that are not reliable under dataset shift. We propose a Bayesian framework to obtain reliable uncertainty estimates for deep classifiers. Our approach consists of a plug-in "generator" used to augment the data with an additional class of points that lie on the boundary of the training data, followed by Bayesian inference on top of features that are trained to distinguish these "out-of-distribution" points.
@article{arxiv.2007.06096,
title = {BaCOUn: Bayesian Classifers with Out-of-Distribution Uncertainty},
author = {Théo Guénais and Dimitris Vamvourellis and Yaniv Yacoby and Finale Doshi-Velez and Weiwei Pan},
journal= {arXiv preprint arXiv:2007.06096},
year = {2020}
}
Comments
ICML 2020 Workshop on Uncertainty and Robustness in Deep Learning