the setting of use_inverted_noise

[string-ellipses]

When I set use_inverted_noise =True, I found that the evaluation metrics on the validation set are quite high even at the begining of the training phase.
Does this potentially mask the true generative performance of the model?