question about argumentation

[siru-zhang]

I am wondering why there is no parameter to control the probability of argumentation. Does this mean that the discriminator has not seen the real data (non-argumented image)?