Название: Differential Privacy Автор: Simson L. Garfinkel Издательство: The MIT Press Серия: The MIT Press Essential Knowledge series Год: 2025 Страниц: 246 Язык: английский Формат: pdf (true) Размер: 10.2 MB
A robust yet accessible introduction to the idea, history, and key applications of differential privacy—the gold standard of algorithmic privacy protection.
Differential privacy (DP) is an increasingly popular, though controversial, approach to protecting personal data. DP protects confidential data by introducing carefully calibrated random numbers, called statistical noise, when the data is used. Google, Apple, and Microsoft have all integrated the technology into their software, and the US Census Bureau used DP to protect data collected in the 2020 census. In this book, Simson Garfinkel presents the underlying ideas of DP, and helps explain why DP is needed in today's information-rich environment, why it was used as the privacy protection mechanism for the 2020 census, and why it is so controversial in some communities.
When DP is used to protect confidential data, like an advertising profile based on the web pages you have viewed with a web browser, the noise makes it impossible for someone to take that profile and reverse engineer, with absolute certainty, the underlying confidential data on which the profile was computed. The book also chronicles the history of DP and describes the key participants and its limitations. Along the way, it also presents a short history of the US Census and other approaches for data protection such as de-identification and k-anonymity.
Protecting confidentiality is especially difficult given the push for agencies to publish microdata, a term commonly used to refer to the individual data records about people, establishments, or some other kind of “unit” that are collected, edited, tabulated, and otherwise used for statistics making. With microdata, researchers outside the statistical agencies can perform their own analyses and even combine data from multiple agencies. In practice, agencies remove the names and other highly identifying attributes before they publish microdata. The danger of publishing microdata is that it is sometimes possible to reidentify the de-identified data. The mosaic effect makes it much easier to mount this kind of attack on public microdata. The world of official statistics uses the term data intruder to refer to a person who performs such an attack, but I prefer the term data hacker.
Навигация
Вход на сайт
Реклама
Differential Privacy 
