Normalizing Flow Coupling Layer

machine learninggenerative modelingprobabilitystatisticscetztikz

Architecture of a normalizing flow coupling layer showing forward and inverse passes. The input is split into two parts: one passes through unchanged while the other is transformed by a learned function conditioned on the first part. Inspired by Ari Seff in https://youtu.be/i7LjDvsLWCg?t=250.