Evaluate training data approach

[juripetersen]

One possible misconception that leads to improper performance in the BVAE could be the quality or the amount of training data we use during training.

As guidance, I use the following papers:

• Understanding disentangling in β-VAE
• Tutorial on VAEs
• Tutorial: Deriving the Standard Variational
  Autoencoder (VAE) Loss Function

I will evaluate our approach with literature once again, making sure that we are not misunderstanding a crucial part of the training for VAEs.