Private Aggregation with Application to Distributed Optimization

The paper presents a fully distributed private aggregation protocol that can be employed in dynamical networks where communication is only assumed on a neighbor-to-neighbor basis. The novelty of the scheme is its low overhead in communication and computation due to a pre-processing phase that can be executed even before the participants know their input to aggregation. Moreover, the scheme is resilient to node drop-outs, and it is defined without introducing any trusted or untrusted third parties. We prove the privacy of the scheme itself and subsequently, we discuss the privacy leakage caused by the output of the scheme. Finally, we discuss implementation of the proposed protocol to solve distributed optimization problems using two versions of the alternating direction method of multipliers (ADMM).