Adversarial Auto Encoder (AAE)
Combination of VAE and GAN
In this article, I’ll be explaining about Adversarial Auto Encoder (AAE), a hybrid between VAE and GAN for generative modelling. Before reading this, I recommend you to read my previous article about Variational Autoencoder (VAE) as I will assume the readers have the knowledge of VAE.
Generative Adversarial Network (GAN)
Generative Adversarial Network or GAN is one of the approaches in deep generative modelling that is becoming really popular right now. The main difference between GAN and VAE is that GAN tries to match the pixel level distribution instead of the data distribution and the way it optimizes the model distribution to the true distribution.
How does GAN generate an image? Images can be thought of as just vectors of the pixel value. However, not any random values can be used to have an image of an object. An image of a dog needs to have a certain value for the pixels and they are arranged in a certain way to make it look like a dog. Hence, we can say that the vectors need to…
