The most impressive magic trick that AI has learned in modern times is when he conjures people out of nowhere. And there's no better machine learning magic than Nvidia's.
Nvidia is a company praised most for its impressive graphics cards. But in the machine learning world, it's one of the most ingenious companies that today uses deep learning. A few years ago TNW reported a new Generative Adversary Network (GAN) that developed the company. At that time, it was an amazing example of how powerful deep learning had become.
This was cutting-edge technology less than a year ago. Today you can use it on your phone. Simply point your web browser at "thispersondoesnotexist.com" and voila : Next time your grandmother asks when you want to settle down with someone nice, you can conjure a picture to show them to them.  A GAN is a neural network that divides the workload of an AI into separate parts. One set of algorithms (one generator) attempts to create something – in this case, a human face – while another set (a judge) tries to determine if it is a true or a false image. If the judge finds out that this is a fake, we have two more weeks of AI winter.
That's not true. Make sure you were still with me. If the judge determines that the image is a fake, the generator starts over. Once the judge is deceived, an AI developer reviews the results and determines if the algorithms need to be adjusted. Nvidia did not invent GAN – the GANfather, Google's Ian Goodfellow, did that. But the company seems to perfect it.
Nvidia's recent efforts – you can read the paper here – is not the same old neural network with a fancy web interface. His layers have been improved, adapted, and equipped with robot therapy to increase their self-esteem (the last lie of another). According to the research team:
Motivated by the style transfer literature, we redesign the generator architecture to find new ways to control the image synthesis process. Our generator starts with a learned constant input and adjusts the "style" of the image in each convolution layer based on the latent code, so that the strength of the image features is directly controlled at different scales.
That is, you can push Refresh everything you want, and it spits out the weirdly convincing faces of people who do not exist. Damn, that's scary.
We will delve deeply into the research to learn more about the future of counterfeit people as soon as we have had the chance to refresh our mouth several times.