How does it work and what are GAN Networks?

Generative modeling is an unsupervised learning task in machine learning that basically involves automatically discovering and learning the regularities or patterns in input data. This happens in such a way that the model can be used to generate or produce new examples that could plausibly have been drawn from the original dataset. For example, if you feed the right kind of neural network with a large collection of images and tell it which ones show dogs and which ones don't, it can eventually learn to discriminate on its own which new, unlabeled images are cats and which are not. A GAN network can be thought of as a pair of competing neural networks: a generator G and a discriminator D. The generator receives as input random noise sampled from some distribution and attempts, thereby, to generate new data intended to resemble real data. The discriminator network tries to discern real data from generated data. As the discriminator network improves its ability to correctly classify the data, metadata is sent back (or “backpropagated”) to the generator network to help it do a better job of trying to fool the discriminator network. Deep learning pioneer and Facebook's Director of Artificial Intelligence, Yann LeCun, even went on to say that GAN Networks were “the coolest idea in deep learning in the last 20 years.” Some researchers from universities and corporations around the world have started creating variations of GANs that perform new classes of tasks. For example, conditional GANs provide additional information to a generator and its discriminator partner, imposing conditions on the generated image. In the following examples provided in a paper written by researchers at the Berkeley AI Research Laboratory, the generator created a colorized version of an existing image and a night view of a daytime image: Image: What are GANs and how do these networks work?

Source Some of the same researchers from Berkeley also developed a technology called Cycle GANs, in which such generated images can happen in both directions – for example, converting a horse image to look like a zebra and vice-versa, check it out below: Image: What are GANs and how do these networks work?

Source