Contrastive Adversarial Training

[CancanZhang]

Hi,

The paper said "The whole training framework follows the contrastive learning framework" as the paper "Supervised Contrastive Learning", but I cannot find the corresponding loss function in the bppattack.py file.

Could you please point out which part is "Supervised Contrastive Learning"? After I checked the bppattack.py file, I found only the cross-entropy loss of the input ([bad_input, negative_input, input]) and their corresponding label are used.

Did you put the vanilla training file instead of the contrastive adversarial training file? If so, could you please upload a contrastive adversarial training file?

Let me know if I understand it wrong. Thanks!

[ZhentingWang]

Hi,

Thanks for your interest in this work.
You can find the code for the contrastive loss in the following repo:
https://github.com/HobbitLong/SupContrast
We will upload the contrastive training file later and let you know once it is done.

Best,
:)

[CancanZhang]

Thanks!