PhD defense Antoine Urban: Efficient Delegated Secure Multiparty Computation

Jury

• M. Renaud SIRDEY, Directeur de recherche, CEA, Université Paris-Saclay, Reviewer
• M. Chen-Da Liu ZHANG, Professeur, HSLU, Reviewer
• Mme Melek ONEN, Professeur, Eurecom, Examiner
• M Phillipe GABORIT, Professeur, Université de Limoges, Examiner
• M. Matthieu RAMBAUD, Maître de conférences, Télécom Paris, Phd supervisor
• M. Duong Hieu PHAN, Professeur, Télécom Paris, Phd supervisor

Abstract

With the rise of storage and computing services in the cloud, it has become possible to delegate data management to remote infrastructures, allowing users to focus solely on data analysis. This architecture also facilitates the seamless combination of data from various sources to extract valuable insights. However, ensuring the confidentiality of outsourced data remains a critical challenge, hindering many potential use cases involving sensitive information.

Secure multiparty computation (MPC) provides a solution to this problem. It enables a group of n participants, each holding private inputs, to compute a function f over their collective data while preserving the privacy of the inputs and ensuring the accuracy of the results. In this thesis, we adopt the cloud computing paradigm to explore the secure delegation of MPC protocols. Specifically, we examine a framework where a group of input-owners delegates computations to untrusted servers, which handle the bulk of the computational workload while adhering to strict security and confidentiality guarantees.

To meet these needs, we rely on solutions based on fully homomorphic encryption (FHE). This powerful tool enables computations directly on encrypted data, ensuring data confidentiality. Three key requirements guide our approach: limiting the protocol to a fixed number of communication rounds (including one or more initial broadcasts followed by peer-to-peer exchanges); ensuring robustness by guaranteeing that honest participants receive correct results even in the presence of malicious behavior by a minority; and enabling simple delegation of computations without complex preprocessing or excessive computational burden for the input-owners.

To address these requirements, we present two main contributions: (i) the first robust MPC protocol leveraging an efficient FHE scheme based on the RLWE assumption, and (ii) a generic approach for designing MPC protocols using an optimally minimal single initial broadcast. Furthermore, our approach enables efficient evaluation, even in large-scale scenarios involving many participants. These advancements provide practical solutions to the challenges posed by MPC, combining simplicity, efficiency, and robustness. They pave the way for secure and efficient applications in diverse contexts where data security and computational performance are paramount.