Enhancing Secure Computations Via Input Certification

Abstract:

This Ph.D. thesis studies two concretely efficient Secure Multi-Party Computation (MPC) protocols that have enhanced security guarantees via input certification. It shows how to efficiently augment an MPC protocol with general-purpose Zero-Knowledge Proof (ZKP) systems to achieve security against malicious adversaries.

This thesis includes two main contributions. The first contribution is in the secure Two-Party Computation setting. It shows how to use the paradigm introduced by Goldreich, Micali, and Wigderson (GMW) efficiently to compile the passively secure Yao's Garbled Circuit protocol into one with security against a malicious Garbler. It shows, through both theoretical analysis and implementation, that the protocol achieves better efficiency than the state-of-the-art. It concludes that the GMW paradigm, which was considered inefficient for general-purpose protocols, can be efficient these days.

 The second contribution of this thesis is a privacy-preserving protocol for machine learning applications. In particular, the thesis introduces a protocol for secure aggregation for Federated Learning where a coordinator and a set of clients and servers train a global machine learning model over the private training data of the clients. The proposed protocol aims to protect the privacy of the local models computed by the clients. This protocol utilizes input certification to prevent malicious clients, which can potentially provide bogus inputs, from attacking the global model. This protocol is robust against malicious corruption of the clients, malicious corruption of a threshold of the servers, and semi-honest corruption of the coordinator. The semi-honest coordinator may collude with the malicious clients and malicious servers. Also, it is robust against client dropouts and dropouts of a threshold of the servers. The implementation and theoretical analysis of the proposed protocol show better computational efficiency compared to the state-of-the-art. Also, the estimated complexity of the communication shows competitive communication complexity.

Advisor: Prof. Muthuramakrishnan Venkitasubramaniam (Computer Science)

Committee: Prof. Lane Hemaspaandra (Computer Science), Prof. Daniel Stefankovic (Computer Science),

Prof. C. Douglas Haessig (University of Arizona)

Chair: Guarav Sharma(Electrical and Computer Engineering)