Beta-VAE

Beta-VAE increases the KL regularization effect of the classical VAE in order to encourage more disentangled latent dimensions. This can make the independent generative factors of the data more clearly separable. It is important for representation learning, controllable generation, and interpretable latent-space studies. However, too much regularization may reduce reconstruction quality.