Torus-based fully homomorphic encryption in federated learning

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Department

Major/Subject

Mcode

SCI3053

Language

en

Pages

82

Series

Abstract

Homomorphic encryption is a cryptographic technique that enables computation on encrypted data, such that the results are as if the computation were performed on the plaintext data. This is all done without decryption at any point, preserving data privacy and security. Some of the main applications of homomorphic encryption include secure data analysis, confidential cloud computing, and private machine learning. For this thesis, a use case was presented by the research institution VTT, in which homomorphic encryption scheme was desired. The scheme presented in this thesis is centered around Torus-based Fully Homomorphic Encryption (TFHE), deriving its naming from a mathematical structure called the torus. The thesis covers the necessary background knowledge in both cryptography and federated learning, as well as the recent developments in TFHE. Noise accumulation in ciphertexts and other limitations of TFHE schemes are discussed. The security of the underlying lattice problems of homomorphic encryption schemes are also briefly analyzed. Finally, a tailored TFHE scheme is then presented for the use case, including a thorough algorithmic construction.

Description

Supervisor

Hollanti, Camilla

Thesis advisor

Bolanos, Wilmar
Vallivaara , Visa

Other note

Citation