Facebook’s new AI teaches itself to see with less human help
Mar 7, 2021
Facebook has recently built an algorithm that can recognize objects in images with little human help. The algorithm, called Seer, was trained on more than a billion images that the company pulled from Instagram. Images with whiskers, fur, and pointy ears, for example, were collected into a pile. The algorithm was then given a small number of human-labeled images -- some of "cats" -- and, lo and behold, it was able to recognize the initial images as belonging to cats or not. So what's the point? Tech like this could be used to read medical images without the need for labeling scans and x-rays. It could be used at Facebook to match ads to posts or help filter out undesirable content. It could even help auto-generate hashtags for Instagram posts.