why do you use abnormal data to train?

[YohannaYin]

when I want to use my self dataset,I found your model just use abnormal dataset of KDD99 to train.But When I read some other paper, they said when they use GAN model to do abnormal detection ,they just use normal dataset to train,not abnormal dataset. Could you answer my question?

[Minqi824]

@YohannaYin . Great question and I agree with you. Normal data are used for constructing the architecture so that the model cannot correctly reconstructed the anomaly samples and thus generates a relatively higher loss (anomaly score). Hope the author could answer this question. @houssamzenati @bruno-31

[Minqi824]

@YohannaYin . I notice that there is a sentence in the paper "Adversarially Learned Anomaly Detection Houssam": "Due to the high proportion of outliers in the KDD dataset, “normal” data are treated as anomalies and the 20% of samples with the highest anomaly scores A(x) are classified as anomalies (positive class)."

Therefore the anomaly samples are treated as normal samples and feeded into the model. Should this be the answer?

[houssamzenati]

Apologies for answering so late and many thanks @jmq19950824 for your reply, indeed we follow the setup of the DAGMM paper so it is as you said