*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
As multi-agent systems grow and become increasingly data-driven, more and more personal data can be shared with unknown or unintended recipients. For example, self-driving cars may share position information for collision avoidance, and smart power grids may share power consumption data to optimize power generation. Even seemingly innocuous data can be very revealing about users, and new data-driven technologies must therefore protect sensitive user data while still allowing networks of agents to function. To address this need, I will present a differentially private implementation for multi-agent tracking control. This talk will use the classic linear-quadratic (LQ) tracking problem to give a broadly applicable problem formulation, and and I will cover a recent privacy implementation that integrates a centralized cloud computer into an otherwise decentralized network. The agents add noise to all data sent to the cloud in order to enforce differential privacy, which gives each agent strong, rigorous privacy guarantees. In contrast to some existing approaches, the cloud does not need to be trusted and instead receives only private information from users, which it then uses to generate control values for them. Functions of private data are therefore fed back into the system. To characterize privacy in feedback, I will present numerical bounds on how difficult it is to compute control values using private user data. The end result of this work is a privacy implementation coupled with the means of quantitatively trading off individuals' privacy and aggregate performance in networks.
