Privacy-preserved signal processing by means of fully homomorphic encryption

In this project, we will revisit the Fully Homomorphic Encryption (FHE) proposed by Gentry and try to design its variant for some practical application of signal processing with sensitive data. Since FHE can perform arbitrary computation without interaction among participant parties while keeping their sensitive data confidential, it is a “holy grail” to solve any Secure Multi-party Computation (SMC) problems theoretically. However, the prohibitive computation complexity and memory requirement make impossible the practical application of FHE. The purpose of this project is to investigate the possibility of using FHE for signal processing in the encrypted domain. We Need to access HPC as typical desktop can run only up 64-bit while we are studying scenarios with security parameters up to 4096-bit. We will compare the complexity of FHE with additively homomorphic Encryption, garbled circuit, and oblivious transfer in some basic algebraic operations (Not, Xor, +, X,), then upgrade the comparison to higher-level operations (Compare, Max, Count) and finally the distance metric (Euclidean, Hamming, or earth mover distance, etc). Based on the implementation results, we will analyze the advantage and disadvantage of all SMC primitives and make a suggestion of the selection of these primitives based on the different signal processing problems.

Computational Methods

We are developing the algorithms on signal processing in the encrypted domain using several cryptographic primitives especially FHE.

Students and Staff

Project participants:
Dr. Sen-Ching Samson Cheung (Advisor)
Ying Luo (Postdoc)

Software

1) Matlab
2) gcc & G++
3) HElib (a software library that implements FHE): https://github.com/shaih/HElib
4) NTL mathematical library (version 6.1.0 or higher): http://www.shoup.net/ntl/
5) GMP (GNU Multi-Precision library): https://gmplib.org/