Building Privacy-Preserving Systems from Communication-Efficient Secure Multi-Party Computation Protocols

Abstract

Enabling collaborative analysis of sensitive distributed data while rigorously preserving privacy represents a central challenge and critical imperative in the modern digital era. Recent progress across domains—ranging from scientific discovery and healthcare to finance and governance—increasingly depends on the ability to extract insights from data held by mutually distrusting parties without compromising individual privacy. Secure multi-party computation (MPC) offers a foundational cryptographic solution to this challenge, allowing joint computation over private inputs without revealing them. Yet despite its strong theoretical underpinnings, MPC's deployment in real-world settings remains limited. The core difficulty lies not only in its inherent communication complexity but also in the diverse, domain-specific challenges that arise when building privacy-assured services in practice. These include aligning cryptographic guarantees with application needs, adapting to heterogeneous trust and deployment models, and ensuring that security, privacy, and efficiency can scale together in operational environments.

This dissertation addresses these barriers by designing and implementing a suite of communication-efficient MPC protocols that enable scalable and robust privacy-preserving computation. It advances both offline preprocessing and online execution with a focus on practical deployability.

The first contribution, HPCG, introduces a programmable and verifiable silent preprocessing architecture grounded in minimal trusted hardware. By generating arbitrary correlated randomness with negligible communication, HPCG decouples the cost of preprocessing from online execution. It ensures correctness and soundness under malicious behavior, and forms a general foundation for fast and lean MPC with verifiable setup—an essential requirement for trust-minimized computation.

The second contribution, Gyges, designs a privacy-preserving broadcasting system for accountable anonymous communication. It combines secret-shared shuffling with an information-theoretic tracing primitive, enabling fine-grained attribution of misuse without compromising the anonymity of honest users. This addresses a fundamental tension in privacy infrastructure, offering a principled solution to enforcing accountability in adversarial environments.

The third contribution, Doppio, enables scalable federated analytics with differential privacy (DP). It proposes an augmented multi-party shuffle DP protocol that replaces the traditional trusted shuffler with a distributed MPC-based mechanism. This approach not only decentralizes trust but also mitigates user-level poisoning attacks and enables flexible trade-offs between privacy, accuracy, and efficiency. The system establishes a new paradigm for structurally robust privacy amplification in practice.

Together, these contributions form a cohesive architecture that advances the practical frontiers of secure computation. By enabling verifiable preprocessing and optimizing MPC for complex privacy properties such as accountable anonymity and differential privacy, this work significantly narrows the gap between cryptographic theory and real-world deployment, paving the way for trustworthy data analytics at scale.

Date of Award	31 Jul 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Cong WANG (Supervisor)