Apple has announced a new technique to improve the performance of its artificial intelligence (AI) Models by taking advantage of user data without compromising the confidentiality of users. In a detailed article on his search blog on automatic learning, the technology giant described how he will use differential confidentiality And synthetic data To train and refine his AI toolsIncluding Genmoji, writing tools and summary features by e-mail.
Apple’s efforts have recently been criticized for disappointing results, especially in areas such as notification summaries and text generation. The main challenge identified by the organization is its ethical standards concerning the use of data, which prevent their models from accessing real content generated by humans. In order to deal with this obstacle, Apple has designed an innovative method that allows modeling of models thanks to methods based on devices while maintaining the exclusivity of data on user devices.
“The synthetic data is created to imitate the format and the important properties of user data, but do not contain any content generated by the user”, “ The company wrote in a blog article. “To organize a representative set of synthetic emails, we start by creating a large set of synthetic messages on a variety of subjects (…), we then derive a representation, called integration, of each synthetic message which captures some of the key dimensions of the message, material, subject and length.”
How does it work?
Apple Start by generating a large set of synthetic content, such as emails or Genmoji prompts, covering a range of subjects and formats. Communications are converted into integrations which represent data by means of models which preserve element frames, tonal characteristics and text lengths. Users who have agreed to the Apple devices analysis program select Apple devices and transmit encrypted data to the models.
The analytical devices validate the results of synthetic integration against the original user content which is available locally, thus helping Apple to determine which models generate the most precise correspondences without requiring the management of any real user information. The methodology allows Apple to develop its models based on authentic use data while maintaining the confidentiality of its users.
Initially used to improve the performance of Genmoji, this method of preserving confidentiality will also be extended to other Apple Intelligence Characteristics, in particular:
- Image Games Area
- Image baguette
- Creation of memories
- Writing
- Visual intelligence
The generation of texts and the summary of emails both apply the same test method at Apple thanks to the correspondence of patterns on the devices of synthetic content. The company obtains natural user communication models, including language choices and structural preferences, via this method without reading the content.
The company declares that its method of protection of privacy works through theoretical mathematical evidence. The system operates while still detecting unique inputs and blocks any attempt to connect information to private users. Apple provides all the treatment via the analysis of devices exclusively on the device for users which allow this functionality according to the established confidentiality standards of the company.
Thanks to this ad, Apple demonstrates its continuous dedication to improving AI’s capacities while maintaining a complete non-invasion concerning user data. The company aims to improve the responsiveness and intelligence of its features while strengthening the confidence of its user base.