Researchers at the College of Charleston, Rutgers University, and Facebook's artificial intelligence (AI) laboratory have modified a generative adversarial network (GAN) to create an AI program that produces images in unconventional styles.
The new GAN uses two neural networks, each submitting solutions and judging the others in an endless loop until the desired result is reached.
One of the networks creates images and the other, trained on 81,500 paintings, determines if those images would be classified as noteworthy works of art. The discriminator network also was trained to distinguish between different styles of art. The researchers trained the generator network to produce images the discriminator recognizes as art, but does not fall into any of the existing styles.
After the AI produced a series of images, members of the public were asked to judge them beside paintings created by people. The researchers found the AI-produced paintings scored slightly higher than those humans created.
From New Scientist
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA