Relationship between Noise addition and Compression based methods for Differential Privacy in Multi-Agent Systems

A multi-agent system(MAS) consists of autonomous agents making observations and sending the data tuple to a fusion center. For authentication and non-repudiation, agents append private parameters like location information, unique identifiers and this leads to an increased privacy risk. A sanitization function is therefore applied to protect privacy. However, application of a sanitization mechanism on the raw data yields decreased utility levels. Common sanitization methods are mainly noise addition based or compression based. In this paper, we introduce novel definitions for system utility and privacy using Cosine similarity, and derive relationship between the two common methods of sanitization. We then experimentally show the interrelation between the two mechanisms.