Facebook announces artificial intelligence to recognize photographic information and explain what kind of photograph by voice

Uploading photos to Facebook may automatically recognize your face and suggest tagging of the person. This is one of the features of Facebook's Deep Learning System, but it is Facebook's Artificial Intelligence Development TeamFacebook AI Research(FAIR) detects objects contained in the photograph and reads out information in the photograph such as "Baby is brushing his teeth in the bathroom", it can answer by asking questions about the photo, etc. We announced the latest artificial intelligence that we have.

New Milestones in Artificial Intelligence Research | Facebook Newsroom
http://newsroom.fb.com/news/2015/11/new-milestones-in-artificial-intelligence-research/

The official announcement of the photo discrimination system using artificial intelligence will be held at the artificial intelligence conference "NIPS" to be held in December 2015. According to the information published at the present time, it is necessary for machine learning of FAIR artificial intelligenceTraining dataDespite being only one tenth of the conventional one, the speed of the identification process of photos will be 30% higher than before.


In addition, a system that can answer questions about information on photos identified by artificial intelligence called "visual Q & A (VQA)" which combines MemNets and photo identification technology has been constructed, and artificial intelligence actually answers questions from the following movies You can see a demonstration to do.

Visual Question and Answering Demo - Facebook Engineering

A baby's photo is displayed on the smartphone, and when asking "Asking if the baby is in the photograph?" By voice input, artificial intelligence answered "yes".


When asking "Where is the baby?", I answered "I am a bathroom" and I can see that I can identify the information in the photo.


The question "What is your baby doing?" I answered "I brush my teeth" and I understand even the situation when photographs were taken.


The artificial intelligence system has a predictive learning function, and by observing and learning visually obtained information, you can understand "what will happen in the future". In addition, FAIR has developed another AI bot pointing to Go in order to let the artificial intelligence system learn the ability to plan. I made two systems compete over the course of several months, the artificial intelligence system grew to the same level as another published game AI, and got the ability of a human master class. As a result, the artificial intelligence system has succeeded in processing by combining the conventional search-based system and the pattern matching system.

You can easily understand how this artificial intelligence system works on Facebook when you see the following movie.

Teaching machines to see and understand - Facebook Engineering

I see a dog with a small camera.


The cable extending from the camera was connected to the notebook PC, and the dog species of the taken dog was identified and displayed.


A woman holding a smartphone in bed seems to be blind because there is a guide dog at her feet.


However, the smartphone you get is identifying posts on Facebook and outputting it by voice, such as "a picture posted at 22:20 last night, pictures are outdoors, clouds are visible, trees are growing thick" .


Another woman also listens to the voice of "artificial intelligence system" in this picture "There are six children, one of them is a microphone", and I am unexpectedly surprised at the voice.


The woman who smiled and said, "Joey is laughing, the number of likes is 19" is a smile.


This woman is also blind, but he says, "It looks like photos are visible in my head."