A collaborative study between the University of Washington and the Allen Institute for Artificial Intelligence (AIAI) produced the world’s first and most-knowledge-hungry computer program dubbed as Learning Everything about Anything (LEVAN).
LEVAN is a fully automated system that scans as many books and images it can online in order to understand everything and anything visual on just about any concept.
“Major information resources such as dictionaries and encyclopedias are moving toward the direction of showing users visual information because it is easier to comprehend and much faster to browse through concepts,” scientist Santosh Divvala says in a statement. He is a research scientist at AIAI and affiliate scientist at UW’s computer science and engineering.
The thing is, he says, these resources have limited coverage because of being manually curated oftentimes as opposed to this newly created system that needs no human curator or supervision at all.
Others say LEVAN is similar to the image search of Google, only a much-improved version because it does not predict what a user might be fascinated in. Instead, it creates a visual archive of the concepts in as many groupings as it can, providing the user a huge visual feast. Also, its search results do not involve the user in numerous clicking or visiting within pages that have nearly identical-looking images.
“It is all about discovering associations between textual and visual data,” says Ali Farhadi, an assistant professor of computer science and engineering at UW. “The program learns to tightly couple rich sets of phrases with pixels in images. This means that it can recognize instances of specific concepts when it sees them.
The system, however, is still young. Currently, it carries the vocabulary skills of a toddler with only about 175 words in tow. These words, for instance, are walking, breakfast, airline, Obama, cancer, robot and eating. In case the concept does not exist, one can always submit any search concept or term for the program to automatically generate a comprehensive listing of subcategory images related to the concept being searched.
It would take up to 12 hours to get to learn broader concepts or terms, which is why the researchers called the public to send their one-word concepts to the system.
In the future, the researchers are hoping to create an application that will make tagging images for archiving quickly. The system for now will continue as an archive. LEVAN will be presented at the annual conference of the Computer Vision and Pattern Recognition in Columbus, Ohio.
LEVAN was rolled out in March, but researchers continue to work on improving the processing capabilities and speed. The U.S. Office of Naval Research, National Science Foundation and University of Washington funded said research.