基于生成对抗网络的模糊密钥加密通信研究  

Study on fuzzy key encryption based on GAN

作　　者：李西明[1] 吴嘉润 吴少乾 郭玉彬[1] 马莎[1] Li Ximing;Wu Jiarun;Wu Shaoqian;Guo Yubin;Ma Sha(College of Mathematics&Informatics,South China Agricultural University,Guangzhou 510642,China)

机构地区：[1]华南农业大学数学与信息学院,广州510642

出　　处：《计算机应用研究》2020年第6期1779-1781,1793,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61872152,61872409);广东省自然科学基金杰出青年基金资助项目(2014A030306021);广东省特支计划资助项目(2015TQ01X796);广州市珠江科技新星资助项目(201610010037)。

摘　　要：解决模糊密钥加密通信问题,并利用生成对抗网络的方法初步实现对称密钥下的模糊密钥加密通信方案。首先利用神经网络实现两方模糊密钥加密通信,实现16 bit密钥对称加密通信中6 bit密钥差异的模糊密钥加密通信。在此基础上考虑敌手存在的模糊密钥加密通信模型,利用GAN思想对通信双方与敌手进行对抗训练,实现16 bit密钥对称加密通信中4 bit密钥差异的模糊密钥通信,实验所得模型中通信双方可正常通信而敌手在可获取密文情况下无法获取明文信息。实验证明了利用神经网络与生成对抗网络解决模糊密钥加密通信问题的可行性。This paper studied the problem of fuzzy key encrypted communication,and preliminarily realized a fuzzy key encryption communication scheme under symmetric key by GAN. It proposed a two-party fuzzy key encrypted communication scheme with neural network model. And the key difference could be 6 bit for symmetric encryption with 16 bit key. Based on these,it gave a fuzzy key encrypted communication scheme with adversaries. This scheme trained the 2-party of communication in adversarial game with their adversary,and as to symmetric encryption with 16 bit key,4 bit key difference was endurable for the communicators while the advisory could get not more auxiliary information from ciphertext. Experiments show that it is feasible to solve fuzzy key encryption communication problem by using neural network and GAN.

关 键 词：生成对抗网络 模糊密钥加密 批规格化 全连接神经网络 卷积神经网络 

分 类 号：TN918[电子电信—通信与信息系统]