Attack and Defense reimplementation

ain-soph / trojanzoo

TrojanZoo provides a universal pytorch platform to conduct security researches (especially backdoor attacks/defenses) of image classification in deep learning.

https://ain-soph.github.io/trojanzoo

GNU General Public License v3.0

274 stars 62 forks source link

Attack and Defense reimplementation #118

Open ain-soph opened 2 years ago

ain-soph commented 2 years ago

In version 1.1, I'll reimplement existing attacks and defenses to eliminate the difference between trojanzoo and original codes.

Attacks:

[x] BadNet
[x] TrojanNN
[x] IMC
[x] LatentBackdoor
[x] Refool
[x] TrojanNet (Embarrasingly Simple Backdoor)
[ ] BypassEmbed
[ ] CleanLabel
[ ] HiddenTrigger

Defenses:

[x] NeuralCleanse
[x] TABOR
[x] ABS
[x] DeepInspect
[x] NeuronInspect
[x] Strip
[x] Neo
[x] ActivationClustering
[ ] SpectralSignature