Home /  Workshop /  Schedules /  Contextually Private Mechanisms

Contextually Private Mechanisms

Connections Workshop: Mathematics and Computer Science of Market and Mechanism Design September 07, 2023 - September 08, 2023

September 08, 2023 (11:00 AM PDT - 12:00 PM PDT)
Speaker(s): Zoe Hitzig (Harvard University)
Location: SLMath: Eisenbud Auditorium, Online/Virtual
Primary Mathematics Subject Classification
Secondary Mathematics Subject Classification No Secondary AMS MSC
Video
No Consent
No Video Uploaded
Abstract

A mechanism designer uses a dynamic protocol to gradually learn agents' private information. A choice rule is \emph{contextually private} if it can be implemented with a protocol that reveals only the private information necessary for determining the outcome. We characterize contextual privacy through local properties of choice rules and construct maximally contextually private protocols through alternating phases of privacy violation and privacy preservation. If the designer can only query one agent at a time, the first-price auction and serial dictatorship are contextually private, while the second-price auction is not. In maximally contextually private implementations of the second-price auction, the designer trades off privacy violations for the winner and the losers.

Supplements No Notes/Supplements Uploaded