*That's an interesting term of art. Lots better than saying "we stuck some noise in your Apple box to make it harder to spy on you."
https://machinelearning.apple.com/2017/12/06/learning-with-privacy-at-scale.html
(...)
Introduction
Gaining insight into the overall user population is crucial to improving the user experience. The data needed to derive such insights is personal and sensitive, and must be kept private. In addition to privacy concerns, practical deployments of learning systems using this data must also consider resource overhead, computation costs, and communication costs. In this article, we give an overview of a system architecture that combines differential privacy and privacy best practices to learn from a user population.
Differential privacy [2] provides a mathematically rigorous definition of privacy and is one of the strongest guarantees of privacy available. It is rooted in the idea that carefully calibrated noise can mask a user’s data. When many people submit data, the noise that has been added averages out and meaningful information emerges.
Within the differential privacy framework, there are two settings: central and local. In our system, we choose not to collect raw data on the server which is required for central differential privacy; hence, we adopt local differential privacy, which is a superior form of privacy [3]. Local differential privacy has the advantage that the data is randomized before being sent from the device, so the server never sees or receives raw data....
