Secure Multi-party Computation Protocols from a High-Level Programming Language
Name
Sander Siim
Abstract
Secure multi-party computation (SMC) enables privacy-preserving computations on data originating from a number of parties.
In today's digital world, data privacy is increasingly more difficult to provide. With SMC methods like
secret sharing and Yao's garbled circuits, it is possible to build privacy-preserving computational protocols that do not leak confidential inputs
to other parties.
The additive secret sharing scheme is very efficient for algebraic ring operations on fixed bit-length data types. However, it is difficult to
build protocols that require robust bit-level manipulation. Yao's garbled circuits approach, in contrast, works on arbitrary bit-length data
and allows the evaluation of any Boolean function. Combining the two methods, we build a secure hybrid protocol, which provides a general method
for building arbitrary secure computations on bitwise secret-shared data. We are able to generate circuits for the protocol easily by using two state-of-the-art C to circuit
compilers designed for SMC applications --- PCF and CBMC-GC. Our hybrid protocol prototype on the Sharemind privacy-preserving computational platform achieves practical performance
comparable to other recent work. In addition to two-party computations, our prototype provides
the ability to perform a set of diverse computations in a generic SMC computational model.
The hybrid protocol together with the circuit compilers provides a simple and efficient toolchain to build general-purpose
SMC protocols for evaluating any Boolean function.
Graduation Thesis language
English
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Dan Bogdanov, Sven Laur
Defence year
2014