Facebook (NASDAQ: FB) has been dominating the news mill lately with its upcoming Moments app which was not granted for release in the EU due to privacy issues affiliated with the network’s face-recognition technology.

The company has recently released a study from its research team regarding Facebook’s artificial intelligence robot ability to learn the looks of objects in its environment. Then, like humans do, the robot apparently remembers the item’s appearance and is capable of creating an image that displays the object based on what it can recall from its memory.

According to the document, the robot auto-generated 64 x 64 pixel photos that looked real enough to trick nearly half of the people that scrutinized them into thinking that they are actual photographs.


The whole procedure is very simple and divided into two parts: One neural network produces an image based on what it remembers – it picks one at random – and then the second neural network examines the file for realism and enhances it if its needed. The research mentioned that the team is planning to develop the robot thus offering larger-pixel images with higher accuracy.

The social network giant has not yet unveiled its plans for the particular project, but it is clear that the past few months big efforts have been in the face recognition department. It is possible to see the mechanism integrated into the Moments app which can figure out which of your Facebook friends are being displayed in a photo with an incredible accuracy.

Facebook is not the only renowned company to have such an initiative in the works. As The Next Web has correctly pointed out, Google (NASDAQ: GOOG) has been developing a similar program but the results were not that brilliant.



VIAThe Next Web
SOURCEFacebook (PDF)