Homomorphic encryption for machine learning and neural networks using high-throughput CRT evaluation

A method comprising:

• receiving, by a hardware accelerator of a computing device, a ciphertext generated by homomorphic encryption (HE) for evaluation;
• decomposing, by the hardware accelerator, coefficients of the ciphertext into a set of decomposed coefficients;
• multiplying, by the hardware accelerator, the decomposed coefficients using a set of smaller modulus determined based on a larger modulus;
• and
• converting, by the hardware accelerator, results of the multiplying back to an original form corresponding to the larger modulus by performing a reverse Chinese remainder theorem (CRT) transform on the results of multiplying the decomposed coefficients;
• wherein the hardware accelerator having hardware circuitry comprising a set of multiplier circuits, reduction circuits, addition circuits, and modular reduction circuits to perform the receiving, the decomposing, the multiplying, and the converting, the hardware circuitry to improve processing unit utilization and latency of computing operations on homomorphically-encrypted data.

A system comprising:

• a memory;
• and
• a hardware accelerator communicably coupled to the memory, the hardware accelerator to implement a Chinese remainder theorem (CRT) multiplier circuit having hardware circuitry comprising m by m multipliers, 2m to m reduction circuits, a ripple addition accumulator, and a modular reduction circuit, the hardware circuitry to improve processing unit utilization and latency of computing operations on homomorphically-encrypted data, wherein the hardware circuitry to:
• receive, from the memory, a ciphertext generated by homomorphic encryption (HE) for evaluation;
• decompose coefficients of the ciphertext into a set of decomposed coefficients;
• multiply the decomposed coefficients using a set of smaller modulus determined based on a larger modulus;
• and
• convert results of the multiplying back to an original form corresponding to the larger modulus by performing a reverse Chinese remainder theorem (CRT) transform on the results of multiplying the decomposed coefficients.