Anomaly Detection by Recombining Gated Unsupervised Experts

Inspired by mixture-of-experts models and the analysis of the hidden activations of neural networks, we introduce a novel unsupervised anomaly detection method called ARGUE. Current anomaly detection methods struggle when the training data does contain multiple notions of normal. We designed ARGUE as a combination of multiple 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. ARGUE achieves superior detection performance across several domains in a purely data-driven way and is more robust to noisy data sets than other state-of-the-art anomaly detection methods.