Guansong

All these datasets have anomalies and there are class labels. You need to hold out the class label information during training, if your models are unsupervised.

Hi, We aim to show that DevNet works well with anomaly-contaminated **unlabeled** data. DevNet performs better when there is no anomaly contamination. See the paper for more details.

> Hi,Mr Pang > I trained model with default arguments by using the backdoor dataset and get the performance as follows > AUC-ROC: 0.7793, AUC-PR: 0.2663 > average AUC-ROC: 0.7834,...

Hi, we didn't have similar problems, but we may release a PyTorch implementation of DevNet later. Stay tuned!

Hi, we focus on inductive settings where test data is assumed to be unavailable during the training stage.