GANs are one of the most interesting topics in machine learning today. They have been used in a number of problems (and not just to generate MNIST digits!) and performed very well in each case. A GAN (General Adversarial Network) consists of a generator and a discriminator, which compete against each other to produce mind-blowing results. Here, we’ll take a mathematical approach towards understanding the GAN and its loss functions. As the idea behind training a GAN comes from game theory, we’ll have a quick look at the Minimax Optimization Strategy too.
In this article, we’ll explore GANs for colourizing B/W images and also learn the loss functions required for our model. So, get ready for some GANs!
“[GANs are] the most interesting idea in the last ten years in machine learning” - Yann LeCun, Director, Facebook AI
GANs have some of the most amazing applications, like turning a horse into a zebra, as seen below.

#programming #machine-learning #artificial-intelligence #tensorflow #heartbeat