Abstract
Differential Privacy (DP) is one of the best known approaches to protecting personal data while extracting useful statistical information. The central idea is to add random noise to published data, in quantities carefully chosen to preserve both the confidentiality and the usefulness of the data.
The speaker introduced the problem of anonymizing personal data and motivated the use of probabilistic approaches. She presented the classic DP approach, based on centralized data processing, followed by a more recent variant, LDP(Local Differential Privacy), where data noise is generated directly by participants. She then introduced the d-privacy model, which unifies the centralized and distributed models and applies to any metric domain, and showed how it can be applied to geolocation data.
