I am writing about the discrimnator loss for FactorVAE. Should the two values in discr_loss be negative? It seems like the probs_z[:, 0] is the probability that the sample comes from the joint distribution, but the line that I linked would encourage the values in probs_z[:, 0] to go towards 0 instead of 1.
Hello,
I am writing about the discrimnator loss for FactorVAE. Should the two values in
discr_loss
be negative? It seems like theprobs_z[:, 0]
is the probability that the sample comes from the joint distribution, but the line that I linked would encourage the values inprobs_z[:, 0]
to go towards 0 instead of 1.