PriMPSO: A Privacy-Preserving Multiagent Particle Swarm Optimization Algorithm.

Area of Science:

• Artificial Intelligence
• Distributed Computing
• Optimization Algorithms

Background:

• Centralized particle swarm optimization (PSO) has limitations in distributed computing, including single-point-of-failure and inadequate privacy protection for particle data.
• Existing distributed PSO algorithms do not effectively safeguard the private data inherent in each particle, such as traveling routes or neural network parameters.

Purpose of the Study:

• To propose a novel privacy-preserving multiagent PSO algorithm (PriMPSO) that protects individual particle data within a distributed computing framework.
• To enable secure data sharing among agents while maintaining the performance of global optimization tasks.

Main Methods:

• PriMPSO utilizes a multiagent system inspired by secure multiparty computation, where each particle is managed by an independent agent.
• The algorithm incorporates a privacy-preserving exemplar selection mechanism and a triple computation protocol for secure particle updates.

Main Results:

• PriMPSO effectively protects the privacy of each particle's data during the optimization process.
• The algorithm demonstrates uniform convergence performance comparable to existing PSO methods in finding optimal solutions.
• Experiments on benchmark and realistic tasks validate the efficacy and privacy-preserving capabilities of PriMPSO.

Conclusions:

• PriMPSO offers a robust solution for privacy-preserving optimization in distributed environments.
• The developed methods for exemplar selection and particle updating are crucial for maintaining both privacy and performance.
• This algorithm advances the application of PSO in scenarios requiring sensitive data handling.