Computer scans Web for images and to learn 'common sense'

Nov. 20, 2013 at 3:52 PM
share with facebook
share with twitter

PITTSBURGH, Nov. 20 (UPI) -- A computer at a U.S. university has been searching the Internet 24 hours a day for images and doing its best to understand them on its own, researchers say.

Running a program called the Never Ending Image Learner, or NEIL, the computer at Carnegie Mellon University is building a growing visual database, generating data that will further enhance the ability of computers to understand the visual world, the researchers said.

The computer is learning to characterize scenes and recognize attributes -- such as colors, lighting and materials -- with a minimum of human supervision, the university reported Wednesday.

NEIL can make associations between these things to obtain common sense information similar to what people observe -- such as that cars often are found on roads or that buildings tend to be vertical.

"Images are the best way to learn visual properties," Abhinav Gupta, a researcher in the university's Robotics Institute, said. "Images also include a lot of common sense information about the world. People learn this by themselves and, with NEIL, we hope that computers will do so as well."

The NEIL program, running since late July, has analyzed 3 million images.

"What we have learned in the last 5-10 years of computer vision research is that the more data you have, the better computer vision becomes," Gupta said.

From its ceaseless image-gathering, NEIL has identified 1,500 types of objects and 1,200 types of scenes.

It has "connected the dots" to learn 2,500 associations from thousands of instances, the researchers said.

Trending Stories