In this paper we introduced Disentangled Auto Encoder (DAE).

img

This is the way to model the latent space of any generative model that can be equipped with an inversion step according to different attribute that we want the model to generate.