MS-Celeb contained faces of celebrities for the purpose of training machines: an estimated 10 million such photos of about 100,000 people were available on the database. The Financial Times has now reported that Microsoft shut it down without notice. The reason given was that the platform was only intended for academic purposes and that the supervising employee no longer worked for the company. An earlier article by the newspaper was nevertheless probably the actual reason behind the deletion. The article reported that the database contained not only pictures of celebrities, but also of numerous ordinary citizens. Many companies had apparently also been using the database commercially without permission.
Researchers tend to be inventive when searching for suitable mass data for machine learning. Technology Review 2018, for example, reported that Turkish scientists have built up a huge recipe database. The data’s structured nature makes it particularly suitable for training text and image recognition.