My task has the following description:
We condition on the noise vector multiple times throughout the network. One of the most popular ways of doing that is via adaptive batch normalization:
π₯=π₯βπππΎ+π½,πΎ=π(π§), π½=π(π§)
The first part of this operation is a standard batch normalization, but instead of optimizing πΎ and π½ as a vector, we optimize functions π and π , which predict affine parameters from a noise vector π§ . Typically these functions are simple linear mappings.
I donβt know how to do that. I am pretty new to PyTorch. Please, help me.