Answers are grounded in this paper and cite their sources.
Efficient implementation of some privacy-preserving algorithms and applications rely on efficient implementation of homomorphic inversion. For example, a recently proposed homomorphic image filtering algorithm and the privacy-preserving body mass index (BMI) calculations repetitively use homomorphic inversion. In this paper, inspired by Montgomery's trick to perform simultaneous plaintext inversion, we tackle the simultaneous homomorphic inversion problem to compute s inverses simultaneously over ciphertexts. The advantage of Montgomery's trick for plaintext arithmetic is well-known. We first observe that the advantage can quickly vanish when homomorphic encryption is employed because of the increased depth of the circuits. Therefore, we propose three algorithms (Montgomery's trick and two other variants) that reduce the number of homomorphic inversions from s to 1 and that offer different levels of trade-offs between the number of multiplications and the circuit depth. We provide a theoretical complexity analysis of our algorithms and implement them using the CKKS scheme in the OpenFHE library. Our experiments show that, for some cases, the run time of homomorphic s-inversion can be improved up to 35 percent while in some other cases, regular inversion seems to outperform Montgomery-based inversion algorithms.
No takes yet. Share an insight, caveat, or question.
Klamti et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: