CS seminar series presents
Quantifying Privacy Leakage in Multi-Agent Planning
by Michal Štolba from Department of Computer Science, FEL
Wednesday, February 10 at 14:30 in 205.
Multi-agent planning is often motivated by the preservation of private information. Such motivation is not only natural for multi-agent systems, but is one of the main reasons, why multi-agent planning problems cannot be solved centrally. Although the motivation is common in the literature, formal treatment of privacy is mostly missing. An exception is a definition of two extreme concepts, weak and strong privacy.
In our recent work, we build on the definition of these two extremes and propose a quantitative measure of privacy leakage in multi-agent planning. The measure is based on the notion of quantitative information flow well established in the field of secure multi-party computation. After definition of the leakage measure for MA-STRIPS-compatible multi-agent planning, we analyze privacy leakage of three predominant multi-agent planning techniques. Finally, we identify general principles reducing privacy leakage in multi-agent planning.