Paper Explained: DCGAN using Keras based on Chintala et al’s work

Generative Adversarial Networks (or GANs) were first introduced in the seminal paper by Goodfellow in 2014. GANs have a huge potential since they can learn to mimic any kind of data. Christie’s sold a portrait for $432,000 that had been generated by a GAN, based on open-source code written by Robbie Barrat of Stanford.

How does it work?

Usually the purpose of a machine learning system is to take in some input or features and produce a result such as a label (a cat, a translation etc.). This is the discriminative function of a machine learning model. The goal of a GAN, on the other hand, is to take a small piece of input such as noise and produce a complex output such as style transfer to an image or producing a new face.

This is where the distinction from discriminative algorithms comes into play.

To read the rest of the article, click here.