Most artificial intelligence (artificial intelligence) continues to be built on a foundation of human effort. If you look inside an AI algorithm, you will find something constructed with data that were curated and annotated by humans.
Now, Facebook has shown how some AI algorithms can learn to do useful work with far less human assistance. The company created an algorithm that learned to recognize objects in images with little help from the tags.
The Facebook algorithm, called Seer, was fed with over a billion images from Instagram, deciding on its own which objects resemble each other. Then, a small number of labeled images were given to the algorithm. It then was able to recognize the images as the algorithm was trained using thousands of labeled examples of each object.
“The results are impressive,” says Olga Russakovsky, an assistant professor at Princeton University who specializes in artificial intelligence (AI) and computer vision. “Getting self-supervised learning to work is very difficult, and major breakthroughs in this area have important implications for improved visual recognition.”
Russakovsky says it's notable that the Instagram images weren't hand-selected to facilitate independent learning.
The research at Facebook is a milestone for an AI approach known as «self-supervised learning», says the chief scientist of Facebook, Yann LeCun.
LeCun pioneered the approach to machine learning known as deep learning , which involves feeding data into large artificial neural networks. About a decade ago, deep learning emerged as a better way to program machines to do all sorts of useful things, like image classification and speech recognition.
But LeCun says the conventional approach, which requires “training” an algorithm by feeding it lots of data, simply won’t scale. “I’ve been a proponent of this whole idea of self-supervised learning for quite some time,” he says. “In the long run, advances in AI will come from programs that watch videos all day long and learn like children.”
LeCun says that self-supervised learning could have many useful applications, for example learning to read medical images without having to label so many scans and X-rays. He says a similar approach is already being used to automatically generate hashtags for images on Instagram. And he says that Seer’s technology could be used on Facebook to match ads to posts or filter out spam.
Facebook's research relies on modifying deep learning algorithms to make them more efficient and effective. Supervised learning has previously been used to translate text from one language to another, but it has been more difficult to apply to images than to words. LeCun says the research team developed a new way for algorithms to learn to recognize images even when part of the image has been altered.
Facebook will release some of the technology behind Seer, but not the algorithm itself, because it was trained using Instagram user data.
Source of information: wired.com
