The social network goliath has made a giant leap in the Artificial Intelligence (AI) domain through its face verification software. Speaking about it, Taigman, a member of the AI research group, said, “We closely approach human performance!”
Identifying a face is routine for human beings but computers have to be taught this skill. Facebook is working on a project known as DeepFace to improve its facial matching software. During a series of tests carried by the social network’s AI research group whereby two pictures were placed side by side, the software got it right 97.25% of the times. In comparison, for the same exercise, a person will likely get it right 97.53% of the times. And, the figures speak for themselves: Facebook’s face-matching software is acquiring the same “recognizing” capabilities as human beings. In what seems like a matter of time, it will even mimic this human ability perfectly!
The significant progress has been made possible thanks to a technique known as Deep Learning which enables a software to process data like a human brain. A circuit of stimulated neurones are used during the learning process along with a pile of data.
To that end, Facebook has gathered 4.4 million faces from its wealthy network to help its software in improving its ability to detect a particular pattern. Like Tagiman further noted, the current software’s performance outpaces previous ones since mistakes have been reduced by over a quarter.
Further, it is worth noting that facial verification or matching is different from facial recognition. Currently, Facebook’s software cannot guess with sufficient accuracy whom a user should tag when a photo is uploaded but it can only say if two pictures display the same person.
Currently Facebook’s project sits in the research sphere only. A presentation of the project by the team is scheduled for the IEEE Conference on Computer Vision and Pattern Vision in June. Meanwhile, a report has been issued “…to get feedback from the research community,” Tagiman added.
Until now, AI has been a well-liked subject in science-fiction books and movies. We have now, however, entered an era where reality transcends fiction!
Source: MIT Technology review