Scaling Matrix for Coupling Layers

• Open: Pasted image 20241119170534.png
• essentially learning a manifold of the data
• smaller the scaling factor, less important the latent dimension
• limit as an element of S approaches 0, the corresponding latent dimension is removed