Privacy-preserving collaborative optimization

With the rapid growth of computing, storing and networking resources, data is not only collected and stored, but also analyzed by different parties. This creates serious privacy problems while inhibiting the use of such distributed data. In turn, this raises the question of whether it is possible to realize value from distributed data without violating security and privacy concerns. Privacy-preserving data analysis has attracted considerable attention in recent years. Specifically, classification, clustering, association rule mining, outlier detection, regression among others, are securely implemented to analyze data privately held by multiple parties. The basic premise of such secure data analysis is that only the data analysis result can be revealed. As a fundamental problem found in many diverse fields, optimization is the study of problems in which one seeks to minimize or maximize a real function by systematically choosing the values of real or integer variables within an allowed set. Inspired from multiparty data analysis, the ubiquitous collection of data opens even greater opportunities in the optimization problems, applicable to the fields of operations research, computer science and mathematics. Collaborative optimization, when done properly with distributed data from different organizations, can facilitate them to improve the allocation of global resources without compromising on security. The primary goal of this dissertation is to develop privacy-preserving collaborative optimization techniques that would allow organizations to gain the maximum value from local information without (or with limited) information disclosure. While answering this problem, an inherent aim is to solve these fundamental problems underlying privacy-preserving analysis and secure multiparty computation (SMC) while making it more accessible and applicable. In this dissertation, we look at fundamental optimization problems such as linear programming, non-linear programming and some classic NP-hard problems. Particularly, we discuss the potential security and privacy concern in the collaborative formulations of them, which occur in real world, such as logistics and scheduling in supply chain management. To securely solve them, we present efficient privacy-preserving methods along with formal security analysis for the proposed privacy notions. In addition, we identify a potential attack to an earlier work and amend the transformation method with enhanced security guarantee. We also address how game theoretic techniques can be used to solve some of the fundamental incentive problems underlying secure multiparty computation in collaborative optimization. The computation/communication cost analysis and the experimental results demonstrate the feasibility, applicability and scalability of the proposed approaches.Ph. D.Includes bibliographical referencesIncludes vitaby Yuan Hon