Double-Adversarial Activation Anomaly Detection: Adversarial Autoencoders are Anomaly Generators

  • 2021-01-12 18:07:34
  • J. -P. Schulze, P. Sperl, K. Böttinger
  • 0


Anomaly detection is a challenging task for machine learning algorithms dueto the inherent class imbalance. It is costly and time-demanding to manuallyanalyse the observed data, thus usually only few known anomalies if any areavailable. Inspired by generative models and the analysis of the hiddenactivations of neural networks, we introduce a novel unsupervised anomalydetection method called DA3D. Here, we use adversarial autoencoders to generateanomalous counterexamples based on the normal data only. These artificialanomalies used during training allow the detection of real, yet unseenanomalies. With our novel generative approach, we transform the unsupervisedtask of anomaly detection to a supervised one, which is more tractable bymachine learning and especially deep learning methods. DA3D surpasses theperformance of state-of-the-art anomaly detection methods in a purelydata-driven way, where no domain knowledge is required.


Quick Read (beta)

loading the full paper ...