Large-Plaintext Functional Bootstrapping in FHE with Small Bootstrapping Keys

TL;DR

This paper introduces a novel polynomial vector encoding for look-up tables in FHE, significantly reducing bootstrapping key sizes and improving efficiency for large plaintexts.

Contribution

It proposes a new polynomial vector encoding method for look-up tables, enabling smaller bootstrapping keys and faster homomorphic evaluation in FHE schemes.

Findings

Reduced bootstrapping key size by polynomial factor

Improved run-time efficiency in homomorphic operations

Enhanced scalability for large plaintexts

Abstract

Functional bootstrapping is a core technique in Fully Homomorphic Encryption (FHE). For large plaintext, to evaluate a general function homomorphically over a ciphertext, in the FHEW/TFHE approach, since the function in look-up table form is encoded in the coefficients of a test polynomial, the degree of the polynomial must be high enough to hold the entire table. This increases the bootstrapping time complexity and memory cost, as the size of bootstrapping keys and keyswitching keys need to be large accordingly. In this paper, we propose to encode the look-up table of any function in a polynomial vector, whose coefficients can hold more data. The corresponding representation of the additive group Zq used in the RGSW-based bootstrapping is the group of monic monomial permutation matrices, which integrates the permutation matrix representation used by Alperin-Sheriff and Peikert in 2014, and the monic monomial representation used in the FHEW/TFHE scheme. We make comprehensive investigation of the new representation, and propose a new bootstrapping algorithm based on it. The new algorithm has the prominent benefit of small bootstrapping key size and small key-switching key size, which leads to polynomial factor improvement in key size, in addition to constant factor improvement in run-time cost.