Autoencoders can be really powerful in finding rich latent spaces. They are almost magical, right? What if we told you that variational autoencoders (VAEs) are even more impressive? Well, they are. They have inherited all the nice things about traditional autoencoders and added the ability to generate data from a parametric distribution.
In this chapter, we will introduce the philosophy behind generative models in the unsupervised deep learning field and their importance in the production of new data. We will present the VAE as a better alternative to a deep autoencoder. At the end of this chapter, you will know where VAEs come from and what their purpose is. You will be able to see the difference between deep and shallow VAE models and you will be able to appreciate the generative property of VAEs.
The chapter is organized as follows:
- Introducing deep...
