同态加密在加密机器学习中的应用研究综述

现有的机器学习算法不能对加密后的数据进行分析计算,而很多领域如医疗、金融等又要求数据保持机密性和安全性,这促进了加密机器学习的产生和发展。同态加密技术是解决这一问题的主要思路,它可以保证在不解密的情况下对密文进行计算,使得解密后的结果与对明文执行相同计算得到的结果相同。文中对同态加密在加密机器学习中的 相关 应用研究进行了综述,主要介绍了目前用同态加密实现加密机器学习的3种算法(加密神经网络、加密k-NN、加密决策树和完全随机森林),并从正确性、安全性、执行效率方面分析了方案设计,总结并对比了不同加密机器学习算法的构造思路,指出了同态加密用于加密机器学习的关键问题和进一步研究需要关注的内容,为同态加密和加密机器学习提供参考。

Survey on Application of Homomorphic Encryption in Encrypted Machine Learning

Nowadays,the existing machine learning algorithms can not analyze and calculate the encrypted data,at the same time,many areas(such as medical industry,financial industry) strongly require data to keep private and secure while analyzed and calculated by untrusted person or company.All these lead to the generation and development of encrypted machine learning.Homomorphic encryption is the primary idea of solving this problem by ensuring that calculations on the cipher text without decrypting,which can result in the same result of the same calculations on the plain text.This paper conducted a survey on application of homomorphic encryption in encrypted machine learning.This work mainly introduced three kinds of algorithms(encrypted neural network,encrypted K-nearest Neighbor,encrypted decision tree and completely random forest) which are used to realize encrypted machine learning with homomorphic encryption,and also analyzed the scheme design from the aspects of correctness,security and efficiency.This paper summarized the construction of different encrypted machine learning algorithms,pointed out the key problems of homomorphic encryption for encrypted machine learning and the content that needs to be focused on in further studies,and provided some referen-ces for homomorphic encryption and encrypted machine learning.