Researchers at Japan's Tokyo Institute of Technology have updated their SOINN machine learning algorithm so that it can now use the Internet to identify items it's never encountered before.
SOINN has been in development for some time now. A few years ago, lead researcher Osamu Hasegawa and his team built a robot that can learn, think, and act by itself. But the new iteration of SOINN has a resource that its predecessor did not: The World Wide Web.
As Diginfo reports, SOINN can now scan objects placed in front of it, like a key, knife or pen. Once given an initial keyword to identify it, the system uses that info to scour an online image repository as it seeks out comparable structures.
After its learning session, SOINN is able to recognize any version of the object it encounters.
This might not sound like much, but it blows away standard facial recognition technology. Hasegawa explains:
With previous methods, for example, face recognition by digital cameras, it's necessary to teach the system quite a lot of things about faces. 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.
Subsequently, SOINN can tell the difference between a box cutter and a knife, or a rickshaw and a car. It's limited to identifying objects in images, but future versions will be able to scan and match content in video and audio.
Image: Tokyo Institute of Technology.