In terms of criticism of the underlying performance of its AI products, especially in territory like this Notification Summary, Apple on Monday Detail Personally analyzing user data with the help of a synthetic data is trying to improve its AI models.
Using a approach ”Differential confidentiality“The agency says it will first produce synthetic data and then compares how accurate its models are with synthetic data snipats generated by the devices (if they have chosen to share device analytics with Apple) and subsequently improve them.
The company wrote in a blog post, “Synthetic data is created to duplicate the user’s data format and important features, but no actual user has produced content,” the company wrote in a blog post. “To correct a representative set of synthetic emails, we start by creating a large set of synthetic messages on different subjects […] We then represent a synthetic message called Embading that captures some of the original dimensions of the message, subject and length. ”
The agency says that these embeddes are then shipped to a number of users selected on device analytics, and the devices are then the most accurate embeddings comparing them with a sample of emails to inform Apple.
The agency says it is using this method to improve its genome models and will use synthetic data for visual intelligence as well as visual intelligence in the future image playground, image wand, memory. Apple says it will also survey users who choose to share device analyzes with synthetic data to improve email shorters.