Learning latent representation with limited labels for IoT anomaly detection

In this paper, we propose an enhanced method (named FeaWAD*) that improves the data encoding strategy based on the FeaWAD network. These models require only a small fraction of anomalies for training.