Privacy preserving solution of DCOPs by mediation

Abstract

In this study, we propose a new paradigm for solving DCOPs, whereby the agents delegate the computational task to a set of external mediators who perform the computations for them in an oblivious manner. That is, the mediators perfectly simulate the operation of the chosen DCOP algorithm, but without getting access to the problem inputs or to its outputs. Specifically, we propose MD-Max-Sum, a mediated implementation of the Max-Sum algorithm. MD-Max-Sum offers topology, constraint, and decision privacy. Moreover, MD-Max-Sum is collusion-secure, as long as the set of mediators has an honest majority. We evaluate the performance of MD-Max-Sum on different benchmarks, problem sizes, and constraint densities. In particular, we compare its performance to PC-SyncBB, the only privacy-preserving DCOP algorithm to date that is collusion-secure, and show the significant advantages of MD-Max-Sum in terms of runtime. We conclude that MD-Max-Sum can be used in practice for solving DCOPs when strong privacy guarantees are required. The main takeaway from this study is a demonstration of the power of mediated computing. It allows either a single party or a set of parties, who may have limited computational or communication resources, to delegate an intricate computation to external dedicated servers who can perform the computation for them in an oblivious manner that protects the privacy of the initiating parties.