Paillier-Numpy

Introduction

​ This project implements the partially homomorphic Paillier algorithm and extends and optimizes encryption, decryption, matrix multiplication, and other operations on Numpy matrices, thereby enhancing speed.

​ Main reference project：https://github.com/data61/python-paillier

​ If there is a need for distributed computing, this project contains a Paillier implementation that I rewrote using Dask. It can perform distributed encryption, decryption, and computation using Dask's distribute functionality. However, it is worth noting that this is merely an experimental project.

Setup

pip install paillier_numpy

How to use

Encrypt and decrypt

import numpy as np
from paillier import generate_paillier_keypair

# create public key and private key
p,q = generate_paillier_keypair()

# create random float
data = np.random.random_sample((100, 100))

# encrypt
encry = p.encrypt(data)

# decrypt
decry = q.decrypt(encry)

Ciphertext addition and subtraction

p, q = generate_paillier_keypair()
data = np.random.random((2, 2))
encry = p.encrypt(data)

# float
encry2 = encry + 3.1415926
print(q.decrypt(encry2) == data + 3.1415926)

# int
encry3 = encry + 666
print(q.decrypt(encry3) == data + 666)

# matrix float
data2 = np.random.random_sample((2, 2))
encry4 = encry + data2
print(q.decrypt(encry4) == data + data2)

# matrix int
data2 = np.random.randint(-100,100,(2, 2))
encry5 = encry + data2
print(q.decrypt(encry5) == data + data2)

Ciphertext Multiplication

p, q = generate_paillier_keypair()
data = np.random.random((2, 2))
encry = p.encrypt(data)

# int
encry2 = encry * 2
print(q.decrypt(encry2)==data*2)

# float
encry2 = encry * 3.141592
print(q.decrypt(encry2)==data*3.141592)

Dot

Note: When performing matrix multiplication, there may be slight errors due to precision issues in the 16th decimal place.

p, q = generate_paillier_keypair()
data = np.random.random((2, 2))
data2 = np.random.random_sample((2,3))
encry = p.encrypt(data)

# [A].dot(B)
encry2 = encry.dot(data2)
print(q.decrypt(encry2)==data.dot(data2))


# B.dot([A])
# !!!!!!!!!!!!!!!!!!!!!!
# [A] must  be converted to Numpy.ndarray
encry3 = data2.T.dot(encry.toArray())
print(q.decrypt(encry3)==data2.T.dot(data))

Multi-process parallelism

Note: The sample code is as follows: where partitions is the number of blocks the matrix is divided into for multiprocessing use.

p, q = generate_paillier_keypair()
data = np.random.random((1000, 100))
data2 = np.random.random_sample((1000,100))

encry = p.encrypt(data)
# time : 0:00:00.624028
decry = q.decrypt(encry)
# time : 0:01:06.076635

encry = p.encrypt(data, is_pool=True, partitions=10)
# time : 0:00:00.282844
decry = q.decrypt(encry, is_pool=True, partitions=10)
# time : 0:00:04.417449

DOT by Multi-process parallelism

Note: Recommend using multi-process matrix multiplication

p, q = generate_paillier_keypair()
data = np.random.random((100, 10))
data2 = np.random.random_sample((10,100))

encry = p.encrypt(data)
encry2 = encry.dot(data2)
# time cost: 0:00:21.922967

encry2 = p.encrypt(data, is_pool=True, partitions=10)
encry3 = encry2.dot(data2,is_pool=True,partitions=10)
# time cost: 0:00:01.600504