Quantcast

Japan's Learning AI Searches the Internet to Identify Objects

By Wesley Fenlon

That's a chair. That's a teapot. That's a glasses?

Japan's Tokyo Institute of Technology has been working on a system called SOINN, or Self-Organizing Incremental Neural Network, for several years. The idea behind SOINN is to develop a learning system that doesn't need predefined inputs to acquire more knowledge. In a recent Diginfo video, researcher Osamu Hasegawa shows off some cool object recognition that's new to SOINN. It looks at objects through a webcam, and after a human-entered keyword, can search the Internet for identifying characteristics of that object. It learns how to identify that thing all by itself, by picking out common characteristics between the scanned object and a simple image search.

With...face recognition by digital cameras, it's necessary to teach the system quite a lot of things about faces," Hasegawa says in the video. "When subjects become diverse, it's very difficult for people to tell the system what sort of characteristics they have, and how many features are sufficient to recognize things. SOINN can pick those features out for itself. It doesn't need models, which is a very big advantage."

Hasegawa also predicts that SOINN will be able to pick identifying characteristics out of video and audio in the future, which would be a big deal for object recognition. Currently, humans entering keywords for SOINN to search objects limit its usefulness, but with a big enough database, the AI could theoretically identify virtually any object.

Check out Diginfo's video with SOINN: