This repository contains the code for the Multitask AutoEncoder Anomaly Detection (MAEAD) method and its implementation as proposed in [1]. MAEAD is a multi-task anomaly detection method for unsupervised data of related tasks. The MAEAD architecture combines concepts from state-of-the-art MTL- and AE-based anomaly detection methods in related problem settings. MAEAD assigns anomaly scores based on the reconstruction performance of a MAE that uses a shared encoding network and a set of task-specific decoder networks.
[1]: Nagelkerken, Sander. Operational Monitoring using Multi-Task Anomaly Detection on Unsupervised Time Series Data, 2023.