About the pre-train and the adv-train

[y52wang]

Hi Wang:
I have a quesion about the Toy_dataset.
In the Generator, it use self.targets to calculate pre-train loss. But, it use self.x to calculate adv-loss.
Why they are different? Any hint will be appreciated.
Thank you very much for your sharing.

[y52wang]

Maybe the answer is hidden in your paper :-)
I'm digging~