Researchers created a program that could allow computers to teach themselves common sense by learning to understand images at a whole new level.
The Never Ending Image Learner (NEIL) will learn to not only recognize elements such as lighting, color, and material; but to make sophisticated connections such as recognizing that a car belongs on a road, a Carnegie Mellon Robotics Institute news release reported.
"Images are the best way to learn visual properties," Abhinav Gupta, assistant research professor in Carnegie Mellon'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."
NEIL has been running since July, and has already analyzed over three million images that encompassed 1,500 objects in half a million images and 1,200 types of scenes out of hundreds of thousands. Overall the program has made 2,500 connections out of thousands of instances.
The researchers hope they will successfully create "the world's largest visual structured knowledge base."
"What we have learned in the last [five to] 10 years of computer vision research is that the more data you have, the better computer vision becomes," Gupta said.
This feat had been attempted in the past by projects such as ImageNet and Visipedia, but the number of images on the web has been too intimidating; there are over 200 billion images just on Facebook.
Since the program is susceptible to errors (like confusing pink the color and Pink the singer) the researchers believe humans are necessary to aid in NEIL's learning process.
"People don't always know how or what to teach computers," Abhinav Shrivastava, a Ph.D. student in robotics, said. "But humans are good at telling computers when they are wrong."
People can also instruct NEIL as to what to search for, but the researchers said the computer program has surprised them with its capabilities.
"As its search proceeds, NEIL develops subcategories of objects - tricycles can be for kids, for adults and can be motorized, or cars come in a variety of brands and models. And it begins to notice associations - that zebras tend to be found in savannahs, for instance, and that stock trading floors are typically crowded," the news release reported.
