Some of the first pieces of common sense a computer ever learned, like the fact that helicopters are found on airfields and babies have eyes, are now available for the public to see online.
Nobody had to tell the Never Ending Image Learner (NEIL) these everyday facts – the computer program figured them out for itself by analysing millions of images from the web all day every day for the last four months.
NEIL uses recent advances in computer vision that allows programs to identify and label objects within images and then recognise elements of them – like colours, lighting and materials – all with the minimum of human supervision. But the project takes things a bit further by trying to make associations between the pictures it sees to come up with everyday rules, such as that cars are often found on roads or that ducks (sort of) look like geese.
Abhinav Gupta, lead investigator and assistant research professor at Carnegie Mellon University’s Robotics Institute, where NEIL spends its time, said that images were the best way to learn visual properties.
“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,” he explained.
Read the rest over on Forbes.
No comments:
Post a Comment