I have trained a VAE that can generate photos of human faces.
I have isolated the dimension that correlates most to smiling and now I only want the VAE to generate smiling faces.
May I know is it a usual practice to adjust the latent_dim value in order to optimize the attributes (e.g. naturalness etc.) of the generated samples?
I suppose an extremely low latent_dim will cause excessive information compression.
How about a high latent_dim? Is it always better to use a higher number of latent_dim?
Or is such a choice of latent_dim a hyperparameter optimization problem that depends on the problem?
p.s. I'm trying to master VAE so GANs is out of context for now.
