Data scientist Alexia Jolicoeur-Marineau has developed the Meow Generator, a collection of machine-learning algorithms that have created more than 15,000 disturbing cat faces.
Jolicoeur-Marineau started with a publicly-available dataset that contains 10,000 picture of cats. She then manually prepared the pictures so just the faces of the cats could be seen, and fed the edited photos into a generative adversarial network (GAN).
GANs have been used on similar projects such as the Massachusetts Institute of Technology's Nightmare Machine, a series of algorithms that generates monster faces and eerie landscapes. GANs consist of two neural networks, one of which creates an image, while the other determines if it looks right.
In the Meow Generator, one of the algorithms abstracts features from the photos to generate its own cat faces. The generated cat faces are then fed into the other algorithm, which determines which images are generated and which are real.
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA