Privacy-preserving parallel computations for graph problems
Date
2023-05-15
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Turvalisel mitmeosalisel arvutusel põhinevate reaalsete privaatsusrakenduste loomine on SMC-protokolli arvutusosaliste ümmarguse keerukuse tõttu keeruline. Privaatsust säilitavate tehnoloogiate uudsuse ja nende probleemidega kaasnevate suurte arvutuskulude tõttu ei ole paralleelseid privaatsust säilitavaid graafikualgoritme veel uuritud. Graafikalgoritmid on paljude arvutiteaduse rakenduste selgroog, nagu navigatsioonisüsteemid, kogukonna tuvastamine, tarneahela võrk, hüperspektraalne kujutis ja hõredad lineaarsed lahendajad. Graafikalgoritmide suurte privaatsete andmekogumite töötlemise kiirendamiseks ja kõrgetasemeliste arvutusnõuete täitmiseks on vaja privaatsust säilitavaid paralleelseid algoritme. Seetõttu esitleb käesolev lõputöö tipptasemel protokolle privaatsuse säilitamise paralleelarvutustes erinevate graafikuprobleemide jaoks, ühe allika lühima tee, kõigi paaride lühima tee, minimaalse ulatuva puu ja metsa ning algebralise tee arvutamise. Need uued protokollid on üles ehitatud kombinatoorsete ja algebraliste graafikualgoritmide põhjal lisaks SMC protokollidele. Nende protokollide koostamiseks kasutatakse ka ühe käsuga mitut andmeoperatsiooni, et vooru keerukust tõhusalt vähendada. Oleme väljapakutud protokollid juurutanud Sharemind SMC platvormil, kasutades erinevaid graafikuid ja võrgukeskkondi. Selles lõputöös kirjeldatakse uudseid paralleelprotokolle koos nendega seotud algoritmide, tulemuste, kiirendamise, hindamiste ja ulatusliku võrdlusuuringuga. Privaatsust säilitavate ühe allika lühimate teede ja minimaalse ulatusega puuprotokollide tegelike juurutuste tulemused näitavad tõhusat meetodit, mis vähendas tööaega võrreldes varasemate töödega sadu kordi. Lisaks ei ole privaatsust säilitavate kõigi paaride lühima tee protokollide hindamine ja ulatuslik võrdlusuuringud sarnased ühegi varasema tööga. Lisaks pole kunagi varem käsitletud privaatsust säilitavaid metsa ja algebralise tee arvutamise protokolle.
Constructing real-world privacy applications based on secure multiparty computation is challenging due to the round complexity of the computation parties of SMC protocol. Due to the novelty of privacy-preserving technologies and the high computational costs associated with these problems, parallel privacy-preserving graph algorithms have not yet been studied. Graph algorithms are the backbone of many applications in computer science, such as navigation systems, community detection, supply chain network, hyperspectral image, and sparse linear solvers. In order to expedite the processing of large private data sets for graphs algorithms and meet high-end computational demands, privacy-preserving parallel algorithms are needed. Therefore, this Thesis presents the state-of-the-art protocols in privacy-preserving parallel computations for different graphs problems, single-source shortest path (SSSP), All-pairs shortest path (APSP), minimum spanning tree (MST) and forest (MSF), and algebraic path computation. These new protocols have been constructed based on combinatorial and algebraic graph algorithms on top of the SMC protocols. Single-instruction-multiple-data (SIMD) operations are also used to build those protocols to reduce the round complexities efficiently. We have implemented the proposed protocols on the Sharemind SMC platform using various graphs and network environments. This Thesis outlines novel parallel protocols with their related algorithms, the results, speed-up, evaluations, and extensive benchmarking. The results of the real implementations of the privacy-preserving single-source shortest paths and minimum spanning tree protocols show an efficient method that reduced the running time hundreds of times compared with previous works. Furthermore, the evaluation and extensive benchmarking of privacy-preserving All-pairs shortest path protocols are not similar to any previous work. Moreover, the privacy-preserving minimum spanning forest and algebraic path computation protocols have never been addressed before.
Constructing real-world privacy applications based on secure multiparty computation is challenging due to the round complexity of the computation parties of SMC protocol. Due to the novelty of privacy-preserving technologies and the high computational costs associated with these problems, parallel privacy-preserving graph algorithms have not yet been studied. Graph algorithms are the backbone of many applications in computer science, such as navigation systems, community detection, supply chain network, hyperspectral image, and sparse linear solvers. In order to expedite the processing of large private data sets for graphs algorithms and meet high-end computational demands, privacy-preserving parallel algorithms are needed. Therefore, this Thesis presents the state-of-the-art protocols in privacy-preserving parallel computations for different graphs problems, single-source shortest path (SSSP), All-pairs shortest path (APSP), minimum spanning tree (MST) and forest (MSF), and algebraic path computation. These new protocols have been constructed based on combinatorial and algebraic graph algorithms on top of the SMC protocols. Single-instruction-multiple-data (SIMD) operations are also used to build those protocols to reduce the round complexities efficiently. We have implemented the proposed protocols on the Sharemind SMC platform using various graphs and network environments. This Thesis outlines novel parallel protocols with their related algorithms, the results, speed-up, evaluations, and extensive benchmarking. The results of the real implementations of the privacy-preserving single-source shortest paths and minimum spanning tree protocols show an efficient method that reduced the running time hundreds of times compared with previous works. Furthermore, the evaluation and extensive benchmarking of privacy-preserving All-pairs shortest path protocols are not similar to any previous work. Moreover, the privacy-preserving minimum spanning forest and algebraic path computation protocols have never been addressed before.
Description
Keywords
secure multi-party computation, graph theory