Short Bytes: Ever wondered how Facebook or Google Photos are able to recognize faces in the photos and identify the spoken commands in the smartphones and even translating them into another language? It’s Artificial Intelligence which uses deep neural networks. Facebook has just come up with an AI called ‘Flow’ which tests and helps in building other AIs.
Back in 2012, Facebook ads team felt the need of an AI engine to improve the company’s ad targeting. The main idea behind the ad targeting was to match ads to almost a billion of people who are using this world’s largest social media platform.
On the other hand, to build such a system, Facebook also had a vast amount of data in building deep neural networks based on the behavior of the users.
According to Facebook, building such engine was not a problem but testing the same posed a challenge. So, they built an AI testing tool called Flow. Hussein Mehanna who is the director of engineering at Facebook and where he founded core machine learning team, a team part of the Applied ML group at Facebook, says
The main designing purpose of the Flow was to help engineers in testing, building and executing the machine learning algorithms which form the backbone of the deep neural networks.
Engineers can also run all sorts of algorithmic possibilities using Flow. According to Mehanna, around 300,000 machine learning models are tested each month at Facebook. However, only one AI model is rolled out into the social network every 60 days. Thus, at the end, Flow is an Artifical Intelligence software which tests and helps in building AI software.
If you are more interested in knowing more about this backbone of Facebook, you can follow this link: FBLearner Flow: Facebook’s AI backbone