Private and Accurate Decentralized Optimization via Encrypted and Structured Functional Perturbation

TL;DR

This paper introduces EFPSN, a decentralized optimization method that ensures privacy of agents' cost functions using encrypted functional perturbations, while maintaining high accuracy and providing theoretical privacy guarantees.

Contribution

The paper presents EFPSN, a novel decentralized optimization algorithm that combines encryption and structured perturbations to achieve privacy without compromising solution accuracy.

Findings

EFPSN guarantees (epsilon, delta)-differential privacy.

Numerical experiments confirm the effectiveness and privacy preservation of EFPSN.

EFPSN can be integrated with existing decentralized algorithms for accurate solutions.

Abstract

We propose a decentralized optimization algorithm that preserves the privacy of agents' cost functions without sacrificing accuracy, termed EFPSN. The algorithm adopts Paillier cryptosystem to construct zero-sum functional perturbations. Then, based on the perturbed cost functions, any existing decentralized optimization algorithm can be utilized to obtain the accurate solution. We theoretically prove that EFPSN is (epsilon, delta)-differentially private and can achieve nearly perfect privacy under deliberate parameter settings. Numerical experiments further confirm the effectiveness of the algorithm.