The AADAE for Unsupervised Anomaly Detection

pytorch == 1.10
torchvision == 0.4.0
numpy == 1.21.5
scipy == 1.4.1
sklearn == 0.0

Publicly available real-world datasets from the ODDS repositor (http://odds.cs.stonybrook.edu)

The experiments are conducted under two task scenarios, Task I is a weakly supervised scenario using only normal samples for training (anomaly-free), and Task II is an unsupervised scenario where the training set is randomly mixed with a few anomalies.

Task I and Task II on Pima

Task I and Task II on Thyroid

Our work is inspired by Gong’s work in Memorizing Normality to Detect Anomaly: Memory-augmented Deep Autoencoder (MemAE) for Unsupervised Anomaly Detection