Deep anomaly detection with self-supervised learning and adversarial training

Highlights：

• Proposing a deep adversarial anomaly detection method to overcome the limitations of existing anomaly detection methods.

• Elaborating an auxiliary classification task to learn the task-specific latent features in a self-supervised learning manner.

• Constructing a deep adversarial training model to do data inference and capture the data distributions in different spaces.

• Designing a majority voting strategy to obtain reliable anomaly detection results.

• Experimental results on several image and sequence datasets show that proposed method outperforms many strong baselines.