GANs, which were introduced by Ian Goodfellow, Yoshua Bengio, and others in NeurIPS 2014, took the world by storm. GANs, which can be applied to all sorts of domains, generate new content or sequences based on the model's learned approximation of real-world data samples. GANs have been used heavily for generating new samples of music and art, such as the faces shown in the following image, none of which existed in the training dataset:
Faces generated by GAN after 60 epochs of training. This image has been taken from https://github.com/gsurma/face_generator.
The amount of realism that's present in the preceding faces demonstrates the power of GANs – they can pretty much learn to generate any sort of pattern when they've been given a good training sample size.Â
The core concept of GANs revolves around the idea of two players...
