Following criticism about the disappointing performance of their AI products, especially in areas such as notification Summaries, Apple on Monday detailed How you are trying to improve your AI models analyzing user data in private with the help of synthetic data.
Using an approach called “Differential privacy“, The company said it would first generate synthetic data and then survey the user devices (provided they have chosen to share the analysis of devices with Apple) with fragments of the synthetic data generated to compare how precise their models are and then improve them.
“Synthetic data is created to imitate the format and important properties of user data, but do not contain any user -generated content,” the company wrote in a blog post. “To cure a representative set of synthetic emails, we begin by creating a large set of synthetic messages in a variety of topics […] Then we derive a representation, called inlaid, from each synthetic message that captures some of the key dimensions of the message, such as language, theme and length. ”
The company said that these inlays are sent to a small number of user devices that have opted for device analysis, and the devices compare them with an electronic email sample to tell Apple what incrustations are more precise.
The company said it is using this approach to improve its Genmoji models, and that in the future it would use synthetic data for the playground, the image wand, the creation of memories and writing tools, as well as visual intelligence. Apple said it would also survey users who choose to share the analysis of devices with synthetic data to improve email summaries.
