International Association
for Cryptologic Research

Lattice-based homomorphic encryption (HE) schemes are based on the noisy encryption technique, where plaintexts are masked with some random noise for security. Recent advanced HE schemes rely on a decomposition technique to manage the growth of noise, which involves a conversion of a ciphertext entry into a short vector followed by multiplication with an evaluation key. Prior to this work, the decomposition procedure turns out to be the most time-consuming part, as it requires discrete Fourier transforms (DFTs) over the base ring for efficient polynomial arithmetic. In this paper, an expensive decomposition operation over a large modulus is replaced with relatively cheap operations over a ring of integers with a small bound. Notably, the cost of DFTs is reduced from quadratic to linear with the level of a ciphertext without any extra noise growth. We demonstrate the implication of our approach by applying it to the key-switching procedure. Our experiments show that the new key-switching method achieves a speedup of 1.2--2.3 or 2.1--3.3 times over the previous method, when the dimension of a base ring is $2^{15}$ or $2^{16}$, respectively.

Computing on data in a manner that preserve the privacy is of growing importance. Multi-Party Computation (MPC) and Homomorphic Encryption (HE) are two cryptographic techniques for privacy-preserving computations. In this work, we have developed efficient UC-secure multiparty protocols for matrix multiplications and two-dimensional convolutions. We built upon the SPDZ framework and integrated the state-of-the-art HE algorithms for matrix multiplication. Our protocol achieved communication cost linear only in the input and output dimensions and not on the number of multiplication operations. We eliminate the ''triple sacrifice'' step of SPDZ to improve efficiency and simplify the zero-knowledge proofs. We implemented our protocols and benchmarked them against the SPDZ LowGear variant (Keller et al. Eurocrypt'18). For multiplying two square matrices of size 128, we reduced the communication cost from 1.54 GB to 12.46 MB, an improvement of over two orders of magnitude that only improves with larger matrix sizes. For evaluating all convolution layers of the ResNet-50 neural network, the communication reduces cost from 5 TB to 41 GB.