基于同态加密和牛顿迭代法的数据隐私保护模型

Data Privacy Protection Model Based on Homomorphic Encryption and Newton Iterative Method

为解决目前机器学习面临的严重隐私泄露问题,提出基于同态加密的大数据隐私保护Logistic回归模型.利用Logistic回归算法对加密数据进行训练和预测,确保在整个过程中不会泄露任何隐私,同时,采用Paillier同态加密算法加密训练数据,利用牛顿迭代算法建立适用于密文数据集的逻辑回归模型.分别在MNIST和Dermatology数据集上执行该模型算法,通过进一步解密明文之后进行评估,进而计算所提模型的准确率.最后,将所提出的模型与相关文献模型进行比较,结果表明:所提模型具有良好的性能和较高的准确率,输出结果与未加密明文运算的输出结果一致,且不影响模型的准确率.所提出的模型可以用于构建二进制分类模型的隐私保护和通过逻辑回归建模的各种问题.

In order to solve the problem of serious privacy leakage faced by machine learning,this paper proposes a Logistic regression model of big data privacy protection based on homomorphic encryption.The model uses Logistic regression algorithm to train and predict encrypted data,and the whole process will not reveal any privacy.The training data is encrypted by using Paillier homomorphic encryption algorithm,and the logistic regression model suitable for ciphertext data set is established by using Newton iterative algorithm.The algorithm can be used as a privacy preserving technique to construct binary classification model,and can be applied to various problems that can be modeled by logistic regression.This paper implements the algorithm on MNIST and dermatology data sets respectively,evaluates the model after further decrypting the plaintext,and then calculates the accuracy of the model.Finally,comparing the model of this paper with the conclusions of relevant literatures,it can be seen that the model of this paper has high accuracy,which shows the feasibility of the model in practical application.