Supervised deep learning methods require a huge amount of data to train models. Acquiring this data is costly and time-consuming. Sometimes, it is impossible to acquire data, as it is not publicly available, or if it is publicly available, the dataset might be very small in size. This is where GANs can come to the rescue. Once trained with a reasonably small dataset, GANs can be deployed to generate new data from the same domain. For example, let's say you are working on an image classification task. You have a dataset, but it is not big enough for your task. We can train a GAN on existing images, and it can then be deployed to generate new images in the same domain. Although GANs currently have training instability problems, several researchers have shown that it is possible to generate realistic images.