Bayesian autoencoders with uncertainty quantification: Towards trustworthy anomaly detection

Despite numerous studies of deep autoencoders (AEs) for unsupervised anomaly detection, AEs still lack a way to express uncertainty in their predictions, crucial ensuring safe and trustworthy machine learning systems high-stake applications. Therefore, this work, the formulation Bayesian (BAEs) is adopted quantify total uncertainty, comprising epistemic aleatoric uncertainties. To evaluate quality we consider task classifying anomalies with additional option rejecting predictions high uncertainty. In addition, use accuracy-rejection curve propose weighted average accuracy as performance metric. Our experiments demonstrate effectiveness BAE on set benchmark datasets two real manufacturing: one condition monitoring, other inspection.