Hi, I have two questions about VAE. Firstly, when training a VAE, we should sample from N(\mu, \sigma^2) to approximate the latent distribution. Should MSE be the average over these samples? Secondly, when testing a VAE, should we also sample from the distribution or just use \mu to get generated picture? Thanks!
Hi, I have two questions about VAE. Firstly, when training a VAE, we should sample from N(\mu, \sigma^2) to approximate the latent distribution. Should MSE be the average over these samples? Secondly, when testing a VAE, should we also sample from the distribution or just use \mu to get generated picture? Thanks!