| tags | |
|---|---|
|
Glow is a flow network introduced by Kingma et al. (2018).
Glow follows up on RealNVP (2016), but creates two new layers:
- actnorm -- learned normalization of activations that replaces Batch Norm used in RealNVP
- 1x1 invertible convolution -- layer that replaces the masking in RealNVP
For more info reference the papers, so far I didn't really need more info 🤷.