Instagram on a phone

3.5bn Instagram pics used to create image recognition system

Image credit: Dreamstime

Facebook has announced that it has been working on an image recognition system to help with moderation, which has been trained on 3.5bn publicly posted photos.

Deep learning is an approach to machine learning often used for image and audio recognition. A deep learning system is approximately inspired by the biological brain, and passes data through a succession of layers, each of which use the previous layer’s output as its input. Deep learning could potentially allow for computers to learn to recognise patterns without the need for such extensive datasets.

In order to teach a machine learning system to recognise various objects, Facebook has harvested 3.5bn images from Instagram, which users have already labelled with 17,000 tags such as #dogs, #avocadotoast or #cockatoo. This means that Facebook employees have not had to spend time hand-labelling the pictures in the dataset; this time-consuming process was described by Mike Schroepfer, Facebook CTO, as the “biggest limiting factor” to making progress with automatic image recognition.

“We rely almost entirely on hand-curated, human-labelled data sets. If a person hasn’t spent the time to label something specific in an image, even the most advanced computer vision systems won’t be able to identify it,” said Schroepfer at F8.

“We built some breakthrough technology that takes publicly available hashtagged images at an unprecedented scale. We have trained on 3.5bn training images using a public set of images without any human curated images in that data set.”

The resulting system is one to two per cent better than any other system according to the ImageNet benchmark, scoring at 84.5 per cent accuracy.

Schroepfer explained that Facebook is already using its more sophisticated image recognition system to help moderate content on the platform, alongside employing an extra 20,000 human moderators to search through reported content to determine what pictures, videos and text is inappropriate, such as that containing nudity, violence, misinformation or hate speech.

“Until very recently, we often had to rely on reactive reports. We had to wait for something bad to be spotted by someone and do something about it,” said Schoepfer. “This is why we are so focused on core [artificial intelligence] research. We require new breakthroughs, and we require new technologies to solve problems all of us want to solve.”

At this year's Facebook F8 conference, Mark Zuckerberg, Facebook CEO, announced a headline-grabbing dating service based on the platform, as well as some other features for Facebook and its subsidiaries.

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them

Close