Facebook researchers are developing advanced facial recognition software, which achieves performance close to that of a human on this specific object.
Specifically, according to a report by MIT Technology Review, when a person is asked if two photos of faces show the same person, the success rate is 97.53%. The software in question is very close, with a rate of 97.25%.
“You don’t normally see this kind of improvement,” says Yaniv Teigman, a member of Facebook’s AI team — a research arm created last year to explore how deep learning (a new approach to artificial intelligence) can help the company. “We’re getting closer to human performance,” he says, noting that the probability of error has been reduced by more than 25 percent compared to previous programs of the same kind.
Facebook's new software, called DeepFace, does what researchers in the field call facial verification (identifying that two images show the same face) rather than facial recognition (matching a face to a name). As for its use for the social network, it would have to do with suggestions for "tags" on users' photos.
Source: koutipandoras.gr
