A new image recognition comes from the hotel’s own research laboratory for artificial intelligence on Facebook. The system recognizes the details on a photo and interpreted the entire scene. Blind people can leave soon “read before a photo”.
At this year’s Web Summit in Dublin Facebook’s Technical Director Mike has presented Sakha a new system, in which one becomes clearer than ever: Facebook is much more than “just” a social media company. What there arose in the Facebook laboratory for artificial intelligence in recent years, has scientific level. The content of images can be detected and linguistically translated using a new algorithm to be implemented in the next few years.
People can read above posts and images with Visual impairment
As an example, Sakha on the Web Summit showed the photo of a skateboard rider on the edge of a halfpipe. The algorithm selected several of the scene matching nouns, verbs and prepositions such as “skateboard, a man, a trick riding, performing, on”. This made the system the set independently: “A man riding a skateboard on ramp”. According to Sakha is the image recognition not only to 30% faster, but requires much less data than comparable solutions.
As concrete applications, Facebook has two things in mind. A better message stream is to provide the General FB-people – after all, 1.5 billion people per month – and Facebook mean by “better”, that the photos on the timeline are specifically selected.
You can get so that to see what you already know or what is like known. Another option would be useful for people with impaired vision. You could have read, of course on Facebook, relevant posts and describe the posted photo.
FAIR: Facebook has a research laboratory for artificial intelligence
The team behind this new algorithms by FAIR – Facebook artificial intelligence research, the only prior was founded two years. For this Facebook chief Mark Zuckerberg took and Mike Sakha some of the best minds in the field of artificial intelligence.
Above all the French Yann LeCun, who runs the FAIR team. He had, so LeCun told a ‘carte blanche’ in a post on Popoular science, sugar mountain, associated with the task “to build the world’s best research laboratory for artificial intelligence”.
The 55-year-old computer specialist, who had made himself a name in the field of artificial neural networks and machine learning, found the an “interesting challenge” and accepted the offer. There are now three teams in different locations with a total of about 30 researchers and 15 engineers. The researchers are approaching the extremely complex field of artificial intelligence from various sites.
The biggest hurdle for artificial intelligence is located in the silent learning
“At the moment the best systems for artificial intelligence are pretty stupid,” says LeCun on popular science. “The biggest hurdle is the unattended learning.” To learn the current systems in two key ways. A method, supervised learning, including the thousands of dogs show the system. It saves the system and from now on recognises special breeds and dogs in General.
In the second type to he learn the computer provided information, the Yes or no to answer to get in his decision making progress. Taking together these two types of learning, that is similar to the, how a person learns. “We observe, draw conclusions and feed everything in our store of knowledge”, says LeCun.
LeCun, stresses the team not by the public wants to seal off his research, on the contrary. “Pretty much all we do is published and a great part of our code is open-source.” The publications are available on the research page of Facebook. “Artificial intelligence is one of the biggest scientific challenges of our time,” LeCun said. “The solutions will come not from a single person, and not by a large company. This will be possible only through a joint effort of all researchers and developers community.”