Bayesian Autoencoders: Analysing and Fixing the Bernoulli likelihood for Out-of-Distribution Detection
Machine Learning
2021-07-29 v1 Machine Learning
Abstract
After an autoencoder (AE) has learnt to reconstruct one dataset, it might be expected that the likelihood on an out-of-distribution (OOD) input would be low. This has been studied as an approach to detect OOD inputs. Recent work showed this intuitive approach can fail for the dataset pairs FashionMNIST vs MNIST. This paper suggests this is due to the use of Bernoulli likelihood and analyses why this is the case, proposing two fixes: 1) Compute the uncertainty of likelihood estimate by using a Bayesian version of the AE. 2) Use alternative distributions to model the likelihood.
Keywords
Cite
@article{arxiv.2107.13304,
title = {Bayesian Autoencoders: Analysing and Fixing the Bernoulli likelihood for Out-of-Distribution Detection},
author = {Bang Xiang Yong and Tim Pearce and Alexandra Brintrup},
journal= {arXiv preprint arXiv:2107.13304},
year = {2021}
}
Comments
Presented at the ICML 2020 Workshop on Uncertainty and Ro-bustness in Deep Learning