Robbie Barrat is testing the limits of machine learning. He gave a neural network lessons in art, and asked it to paint landscapes (digitally, of course). More recently, he turned his attention to classical nudes. He fed thousands of nude portraits into a Generative Adversarial Network (GAN) to see what would result.
Generative adversarial networks are defined as a class of artificial intelligence algorithms used in unsupervised machine learning, which uses two different neural networks, one called the "generator" and one the "discriminator."
"The generator tries to come up with paintings that fool the discriminator, and the discriminator tries to learn how to tell the difference between real paintings from the dataset and fake paintings the generator feeds it," Barrat told me. "They both get better and better at their jobs over time, so the longer the GAN is trained, the more realistic the outputs will be."
Sometimes, Barrat explained, the GAN will fall into what's called a "local minima," which means the generator and discriminator have found a way to keep trying to fool each other without actually getting better at the task at hand.
Here are some AI generated nude portraits I've been working on
— Robbie Barrat (@DrBeef_) March 27, 2018
Usually the machine just paints people as blobs of flesh with tendrils and limbs randomly growing out - I think it's really surreal. I wonder if that's how machines see us... pic.twitter.com/tYgzCHGfse
The two neural networks may be quite pleased with themselves, but their nudes are so abstract that they'll never flag a moderator. The one in the middle right of this Tweet looks like Olaf from Frozen. However, people have asked if they could buy a print. Read about the algorithmically-generated art at CNET and see more digital paintings in Barrat's Twitter feed. -via Boing Boing