The U.S. Defense Advanced Research Projects Agency's Generative Adversarial Networks for Design Exploration and Refinement (GANDER) project has awarded a grant to Pennsylvania State University (PSU) researchers to train computers to produce original design ideas and then see if they are technically feasible.
The approach is unique in that it is a transition from only teaching computers how to classify the difference between things in an environment toward teaching them to be creative.
The PSU team is using generative adversarial networks, which PSU professor Conrad Tucker says involve an idea generator network competing with a discriminator network. They propose creating the latter using simulation environments to embed knowledge about physics and physical properties so generated designs are grounded in solid physical laws.
Tucker says making it possible to substantially reduce the time it takes for an artificial intelligence agent to learn about the physical properties of its surroundings will have enormous potential for autonomous systems overall.
From Penn State News
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA