Generative Adversarial Networks

GANs are networks that generate new, synthetic data by learning patterns and underlying representations from a training dataset. The GAN does this by using two networks that compete with one another in an adversarial fashion. These networks are called the generator and discriminator.

To see how these networks compete with one another, consider the following example. The example will skip over a few details that will make more sense as you get to them later in the chapter.

Imagine two entities: a money counterfeiter and a business owner. The counterfeiter attempts to make a currency that looks authentic to fool the business owner into thinking the currency is legitimate. By contrast, the business owner tries to identify any fake bills, so that they don't end up with just a piece of worthless paper instead of real currency.

This is essentially what GANs do. The counterfeiter in this example is the generator, and the business owner is...