When Ian Goodfellow dreamt up the idea of Generative Adversarial Networks (GANs) over a mug of beer back in 2014, he probably didn't expect to see the field advance so fast: In case you don't see where I'm going here, the images you just saw were utterly, undeniably, 100% … fake. Also, I don't mean these were photoshopped, CGI-ed, or (fill in the blanks with whatever Nvidia's calling their fancy new tech at the moment). I mean that these images are entirely generated through addition, multiplication, and splurging ludicrous amounts of cash on GPU computation. The algorithm that makes is stuff work is called a generative adversarial network (which is the long way of writing GAN, for those of you still stuck in machine learning acronym land), and over the last few years, there have been more innovations dedicated to making it work than there have been privacy scandals at Facebook. Summarizing every single improvement to the 2014 vanilla GANs is about as hard as watching season 8 of Game of Thrones on repeat. I'm not going to explain concepts like transposed convolutions and Wasserstein distance in detail. Instead, I'll provide links to some of the best resources you can use to quickly learn about these concepts so that you can see how they fit into the big picture. If you're still reading, I'm going to assume that you know the basics of deep learning and that you know how convolutional neural networks work.

Generative Adversarial Networks are a powerful class of neural networks with remarkable applications. They essentially consist of a system of two neural networks -- the Generator and the Discriminator -- dueling each other. Given a set of target samples, the Generator tries to produce samples that can fool the Discriminator into believing they are real. The Discriminator tries to resolve real (target) samples from fake (generated) samples. Using this iterative training approach, we eventually end up with a Generator that is really good at generating samples similar to the target samples. GANs have a plethora of applications, as they can learn to mimic data distributions of almost any kind.

Generative Adversarial Networks are a powerful class of neural networks with remarkable applications. They essentially consist of a system of two neural networks -- the Generator and the Discriminator -- dueling each other. Given a set of target samples, the Generator tries to produce samples that can fool the Discriminator into believing they are real. The Discriminator tries to resolve real (target) samples from fake (generated) samples. Using this iterative training approach, we eventually end up with a Generator that is really good at generating samples similar to the target samples. GANs have a plethora of applications, as they can learn to mimic data distributions of almost any kind.

We address the problem of defining a network graph on a large collection of classes. Each class is comprised of a collection of data points, sampled in a non i.i.d. way, from some unknown underlying distribution. The application we consider in this paper is a large scale high dimensional survey of people living in the US, and the question of how similar or different are the various counties in which these people live. We use a co-clustering diffusion metric to learn the underlying distribution of people, and build an approximate earth mover's distance algorithm using this data adaptive transportation cost.