Differential privacy
Differential privacy is a mathematically-rigorous definition of privacy. An algorithm uses a dataset to calculate its output. An algorithm is said to be differentially private if, based on its output, it is impossible to tell whether or not a particular individual was in the dataset.
In simpler terms, this property is fulfilled if the algorithm's behavior does not noticeably change when a single individual joins or leaves the dataset.[1]
Use cases
Differential privacy is used in data collection on mobile devices. Operators can use this data for learning better models. One example is the keyboard data in Android.[2] Another example is the usage data on iPhone[3]
Differential Privacy Media
An informal definition of differential privacy
A formal definition of ε-differential privacy. D_1 is a dataset without the private data, and D_2 is one with it. This is "pure ε-differential privacy", meaning δ=0.
Laplace mechanism offering .5-differential privacy for a function with sensitivity 1.
