Facebook's AI research team 'FAIR' opens a page looking back on its 5th anniversary



Facebook AI researcher group " FAIR (Facebook AI Research)" founded at the end of 2013 celebrates its 5th anniversary in 2018 in birth. A page summarizing FAIR's trajectory that the developed AI technology is utilized also on Facebook is released.

FAIR at 5: Facebook Artificial Intelligence Research accomplishments
https://code.fb.com/ai-research/fair-fifth-anniversary/

In 2014, FAIR announced DeepFace that brings the accuracy of AI's face authentication closer to the same level as humans. Using nine layers of neural networks, we succeeded in raising the accuracy of face recognition to 97.35%.



FAIR technology is said to be available as open source. Movie analysis classifier C3D announced in 2015 incorporates "three-dimensional Convolution" which extends the dimension in the time direction of the analysis of two-dimensional images.



FastText in 2016 is a machine learning library for making natural language processing faster and more accurate, making it possible to analyze and classify various sentences. Of course, it should be used to analyze and classify a huge number of texts posted on Facebook.



FAISS announced in 2017 is an algorithm that enables efficient nearest neighbor search by using GPU for large scale data. It is used for analyzing datasets with more than 1 billion data, and it is used by more than 5000 GitHub users.



In 2018, we released ELF OpenGo as inspired by DeepMind's AlphaGo as an open source. Based on FAIR's ELF platform, learned models and datasets used for learning are released.



In the past 5 years, video analysis technology has also greatly evolved. Detailed recognition up to detailed elements is possible as follows ... ...



It reaches the level which can understand the object included in the moving image in three dimensions.

DensePose: Creating a 3D surface on top of people as they move

A new tool we call DensePose occurrences full 3D world that can be applied, in real time, to footage of human bodies in motion.

Posted by Facebook Engineering Posted on Monday, April 30, 2018


At the time the FIAR was founded, the aim was to make the computer have the same intelligence (intelligence) as a human being. Even if it celebrates the fifth anniversary, its aim has been maintained unchanged, and we have a policy to continue developing higher technology in the future.

in Note,   Software,   Video, Posted by darkhorse_log