Demystifying Inductive Biases for (Beta-)VAE Based Architectures

                 Dominik Zietlow 1 Michal Rolnek 1 Georg Martius 1

             Abstract                strates both high semantic quality as well as good perfor-
   The performance of β-Variational-Autoencoders       mance in terms of disentanglement. Until today, derivates of
   (β-VAEs) and their variants on learning seman-       VAEs (Higgins et al., 2017; Kim & Mnih, 2018a; Chen et al.,
   tically meaningful, disentangled  ...