Using GANs

We looked at GANs and how they are used in model extraction attacks in Chapter 8. GANs were introduced by Ian Goodfellow and his colleagues in 2014. More information can be found at https://dl.acm.org/doi/10.5555/3157096.3157346.

They are used to generate realistic but artificial data such as images and video and have improved enormously over the last 10 years. The following figure, taken from a paper published in 2018, shows the improvements in generated image fidelity over the years:

Figure 11.1 – The progress in the quality of AI-generated images over the years (source: https://arxiv.org/abs/1802.07228)

Applications of GANs include the following:

• Image generation: GANs can generate realistic images. This has applications in fields such as art, where GANs can create new artworks, as well as in entertainment for special effects.
• Data augmentation: In scenarios where data is scarce, GANs can generate additional data for training...