TUM Logo

Anomaly Detection by Recombining Gated Unsupervised Experts

Anomaly detection has been considered under several extents of prior knowledge. Unsupervised methods do not require any labelled data, whereas semi-supervised methods leverage some known anomalies. Inspired by mixture-of-experts models and the analysis of the hidden activations of neural networks, we introduce a novel data-driven anomaly detection method called ARGUE. Our method is not only applicable to unsupervised and semi-supervised environments, but also profits from prior knowledge of self-supervised settings. We designed ARGUE as a combination of dedicated expert networks, which specialise on parts of the input data. For its final decision, ARGUE fuses the distributed knowledge across the expert systems using a gated mixture-of-experts architecture. Our evaluation motivates that prior knowledge about the normal data distribution may be as valuable as known anomalies.

Anomaly Detection by Recombining Gated Unsupervised Experts

International Joint Conference on Neural Networks, IJCNN 2022. Proceedings

Authors: Jan-Philipp Schulze, Philip Sperl, and Konstantin Böttinger
Year/month: 2022/9
Booktitle: International Joint Conference on Neural Networks, IJCNN 2022. Proceedings
Fulltext: click here

Abstract

Anomaly detection has been considered under several extents of prior knowledge. Unsupervised methods do not require any labelled data, whereas semi-supervised methods leverage some known anomalies. Inspired by mixture-of-experts models and the analysis of the hidden activations of neural networks, we introduce a novel data-driven anomaly detection method called ARGUE. Our method is not only applicable to unsupervised and semi-supervised environments, but also profits from prior knowledge of self-supervised settings. We designed ARGUE as a combination of dedicated expert networks, which specialise on parts of the input data. For its final decision, ARGUE fuses the distributed knowledge across the expert systems using a gated mixture-of-experts architecture. Our evaluation motivates that prior knowledge about the normal data distribution may be as valuable as known anomalies.

Bibtex:

@inproceedings {
author = { Jan-Philipp Schulze and Philip Sperl and Konstantin Böttinger},
title = { Anomaly Detection by Recombining Gated Unsupervised Experts },
year = { 2022 },
month = { September },
booktitle = { International Joint Conference on Neural Networks, IJCNN 2022. Proceedings },
url = { https://doi.org/10.1109/IJCNN55064.2022.9892807 },

}