Privacy-preserving multi-objective evolutionary algorithms

Abstract

Existing privacy-preserving evolutionary algorithms are limited to specific problems securing only cost function evaluation. This lack of functionality and security prevents their use for many security sensitive business optimization problems, such as our use case in collaborative supply chain management. We present a technique to construct privacy-preserving algorithms that address multi-objective problems and secure the entire algorithm including survivor selection. We improve performance over Yao's protocol for privacy-preserving algorithms and achieve solution quality only slightly inferior to the multi-objective evolutionary algorithm NSGA-II. © 2010 Springer-Verlag.