Distributed privacy-preserving mean estimation
Mirco Schoenfeld, Martin Werner · 2014
Due to the rise of mobile computing and smartphones, a lot of information about groups has become accessible. This information shall often be kept secret. Hence distributed algorithms for privacy-preserving distribution estimation are needed. Most research currently focuses on privacy in a database, where a single entity has collected the secret information and privacy is ensured between query results and the database. In fully distributed systems such as sensor networks it is often infeasible to move the data towards a central entity for processing. Instead, distributed algorithms are needed. With this paper we propose a fully distributed, privacy-friendly, consensus-based approach. In our approach all nodes cooperate to generate a sufficiently random obfuscation of their secret values until the estimated and obfuscated values of the individual nodes can be safely published. Then the calculations can be done on this replacement containing only non-secret values but recovering some aspects (mean, standard deviation) of the original distribution.