Facebook has built an AI backbone, powered by a 40-PFLOPS GPU cluster, which is behind much of the Facebook experience.
To improve its user experience, Facebook now conducts 50 times more AI experiments per day than a year ago. This helps put research into production faster.
The social network highlighted several ways AI powers various Facebook experiences, as detailed below.
To remove communication barriers, the Applied Machine Learning team built an AI-based automatic translation system that helps 800 million people see translated posts in their News Feed every month.
Using Facebook’s automatic image classifiers, users could search through photos their friends have shared to look for one based on the image content, instead of relying on tags or surrounding text.
Facebook is building systems that can understand images at the individual pixel level, rather than classifying the entire image.
This is called image segmentation, and will allow for more immersive experiences for the visually impaired through “talking images”.
Real-time video classification
Facebook wants to classify live videos in real time, using computer vision techniques to understand and classify what’s in a video.