Efficient Multiparty Fully Homomorphic Encryption With Computation Fairness and Error Detection in Privacy Preserving Multisource Data Mining

• Published in: IEEE transactions on reliability Vol. 72; no. 4; pp. 1308 - 1323
• Main Authors: Guo, Guanglai, Zhu, Yan, Chen, E., Yu, Ruyun, Zhang, Lejun, Lv, Kewei, Feng, Rongquan
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Bivariate analysis; Cluster analysis; Clustering; Computational modeling; Data mining; Data privacy; Degree reduction; Distributed databases; Error correction & detection; Error detection; Homomorphic encryption; homomorphic encryption (HE); Mathematical analysis; Multiplication; multisource data mining; Polynomials; Privacy; privacy preservation; Protocols; Reconstruction; Reliability; secure multiparty computation; Vector quantization
• ISSN: 0018-9529; 1558-1721
• DOI: 10.1109/TR.2023.3246563

In this article, we address the problem of data privacy in multisource data mining. To do it, we present a new multiparty fully homomorphic encryption (MP-FHE) scheme, in which all participants are completely fair to perform the same computation. At first, the proposed MP-FHE scheme is divided into five stages (i.e., calculation, configuration, recombination, resharing, and reconstruction stage) to achieve the unified computation form of addition and multiplication. Meanwhile, random bivariate polynomials and commutative encryption are used to achieve the degree reduction of polynomials and the continuity of computation. Moreover, we prove that the scheme meets result consistency and program termination under the fail-stop adversary model. Especially, three kinds of error detection criteria are presented to find errors in three different stages (i.e., recombination, resharing, and reconstruction stage), which provides the monitor basis for the fail-stop adversary model. In addition, the MP-FHE scheme is applied into privacy preserving k-means clustering algorithm. Finally, we evaluate the computation and communication performance of our scheme from both theoretical and experimental aspects, and the evaluation results show that the scheme is efficient enough for multisource data mining.