Apple details How to plan to improve artificial intelligence models by analyzing user data in particular

Following criticism about the amazing performance of artificial intelligence products, especially in areas such as notice SummariesApple on Monday detailed How to try to improve artificial intelligence models by analyzing user data in particular with the help of artificial data.

Using an approach called “Conversion privacyThe company said that it will first create artificial data and then reconnaissance users (provided that they have chosen to share devices analyzes with Apple) with excerpts from artificial data created to compare the accuracy of their models, thus improving them.

The company wrote in a blog post: “Artificial data is created to imitate the important user data format, but it does not contain any real content of users,” the company wrote in a blog post. “To raise a representative set of artificial emails, we start creating a large collection of artificial messages in a variety of topics (…) We then derive a representation, called inclusion, for every synthetic message that captures some of the main dimensions of the message such as language, subject and length.”

The company said that these implications are then sent to a small number of user devices that chose devices analyzes, then compare the devices with them with a sample of Apple emails that the most accurate implications.

The company said it uses this approach to improve Genmoji models, and in the future it will use the artificial data of the photo stadium, photo juices, memories creation and writing tools as well as visual intelligence. Apple said it will also explore users who choose to share devices analyzes with artificial data to improve email summaries.