Facebook algorithms will be able to recognize "drunken" photos.

Facebook is working on software that will automatically recognize drunk people and block photographs of users in indecent form so that the next morning they will not be painfully ashamed in front of parents and friends. Naturally, in the morning a person can cancel the lock and confirm the publication of photos.

To implement this idea, you need to use pattern recognition algorithms and an artificial intelligence system that can distinguish the sober. The task is really difficult, because even people do not always cope with it. As in the joke about Trudovik .

The creation of such a system is engaged in a researcher from New York University and machine learning gurus Yann LeCun. He heads the laboratory of artificial intelligence , which was created in the company (Facebook Artificial Intelligence Research lab, FAIR).
')
The system is trained using standard neural network learning methods. There is a large database of photos in which people tag photos of sober and drunk people, and the system gradually finds signs that are known only to it, by which these photos can be distinguished.

In addition, Facebook already confidently recognizes faces and identifies people in photographs, so to determine the degree of intoxication, you can use the archive of previous images of the individual, in which he is guaranteed to be sober (that is, which he explicitly allowed for publication).

Facebook is not the only company working in the field of “deep learning” algorithms. This is a very important technology that is used in its activities by Google, Microsoft, Yahoo, Twitter, Baidu and others.

In the long term, Facebook expects to create a system that will understand the meaning of published messages and other content, a kind of “digital assistant”. According to Lekun, in the long term this system can be used in combination with Oculus VR virtual reality technology (which Facebook recently bought) and robots.