search

S-Mohammad-Hashemi / RePO

Enhancing Robustness Against Adversarial Examples in Network Intrusion Detection Systems
16 stars 10 forks source link

good idea to improve reconstruction anomaly #1

Open yue123161 opened 3 years ago

yue123161 commented 3 years ago

It is a good idea to use partial reconstruction error for anomaly detection. As you say in the paper, the RePo is compared with Kitsune, which is based on ensemble of autoencoders. But as i see in Kitsune Paper，they use a sub-space of features as input for each autoencoder. So what is the difference between partial reconstruction and subspace reconstruction.

Best Regards.

S-Mohammad-Hashemi commented 3 years ago

Thank you for your interest in our paper. In Kitsune each autoencoder receives a subset of features as input but it reconstructs the same subset. So essentially each autoencoder would approximate an identity function. Whereas in our work the model is forced to reconstruct the hidden parts of the input based on the visible parts.

yue123161 commented 3 years ago

Thanks for you response Can you share the packet and flow preprocess code in this git

rockersiyuan commented 2 years ago

Your solution is so excellent.Can you share the packet and flow preprocess code in this git？