acm-header
Sign In

Communications of the ACM

ACM TechNews

A Google Intern Built the AI Behind These Shockingly Good Fake Images


View as: Print Mobile App Share:
Realistic fake images.

Google researchers developed an artificial intelligence system called BigGAN that generates incredibly realistic fake images.

Credit: Andrew Brock

Google researchers have developed an artificial intelligence (AI) system that generates incredibly realistic fake images.

They say BigGAN is a generative adversarial network composed of two models: one that creates random images out of random numbers, and another that compares these images to real images and tells the generator how close it is to the real thing.

Although BigGAN is similar to other generative adversarial networks, its main advantage is the sheer amount of computing power that the researchers used; by adding more nodes to increase the complexity of the neural network and showing the model more images than most researchers do, the team was able to create a system that more accurately understands and models textures.

The researchers also used the "truncation trick," which lowers the random numbers that the generator uses to create its images.

From Fast Company
View Full Article

 

Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account