Coalitional Bayesian autoencoders: Towards explainable unsupervised deep learning with applications to condition monitoring under covariate shift

Download PDF
Open Webpage

Bang Xiang Yong, Alexandra Brintrup

Published: 2022, Last Modified: 15 May 2025Appl. Soft Comput. 2022EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Two sensor attribution methods are formulated using the Bayesian Autoencoder (BAE).•Comprehensive quantitative evaluation metrics based on covariate shift.•Conventional BAE is found to suffer from highly correlated misleading explanations.•A novel configuration of “Coalitional BAE” is proposed to improve explanation quality.