Facebook is using Instagram images and hashtags to train its AI

Facebook explained at the F8 developer’s conference that it took billions of public Instagram photos to train its image recognition models.

The social network is collecting images that are tagged with relevant hashtags and using them to train its photo recognition AI. Typically AI is trained by real people annotating photos so that the computer can start to learn what’s in the picture. Facebook is speeding up this process by using pre captioned photos from Instagram to train its AI automatically.

While this version of AI training seems like it wouldn’t be as accurate as a human trained model but Facebook is stating that its AI has an 85.4 percent accuracy rating on ImageNet.

To make sure that the AI is being trained correctly Facebook created a “large-scale hashtag prediction model” that is able to separate the irrelevant hashtags from the specific ones, according to a report from TechCrunch.

This technique helps Facebook tell the difference between dog breeds, types of food and plants.

While this method of training AI seems intrusive on users privacy Facebook only takes images that are publicly posted, so users who have a private account don’t have to worry about there data being scraped.

It will be interesting to see what the social networks end up using all this data for and what kind of features it will bring to the platform.

Source: TechCrunch