生成对抗机制的密码算法组件化设计  

Componential Design of Cryptographic Algorithm Based on Generative Adversarial Method

作　　者：周兴晖 王朝 金鑫[2] ZHOU Xinghui;WANG Zhao;JIN Xin(University of Science and Technology of China,Hefei 230026,P.R.China;Beijing Electronic Science and Technology Institute,Beijing 100070,P.R.China)

机构地区：[1]中国科学技术大学,合肥市230026 [2]北京电子科技学院,北京市100070

出　　处：《北京电子科技学院学报》2020年第4期1-15,共15页Journal of Beijing Electronic Science And Technology Institute

基　　金：国家自然科学基金面上项目,无退化的混沌密码标准实现研究,项目编号:61772047。

摘　　要：将"自动生成对抗"的思想与密码算法设计与分析领域相结合,提出利用生成对抗方法进行组件化密码算法自动设计的方案.首先,利用基于深度神经网络的密码分析构建密码破解器,与密码算法进行自动地对抗竞争学习,从而实现分组密码算法的自动设计.其次,通过分析生成对抗机制,设计生成对抗网络的对称加密模型.最后,通过对不同神经网络的使用与尝试,最大限度地使神经网络设计的分组密码算法在第三方干扰的情况下进行加解密.同时,在基于生成对抗机制的基础上,对原始的对称加密模型进行改进,增强第三方窃密者的攻击性,使得设计出来的加密算法可以抵御不同形式的攻击,从而提升密码算法的安全性.In this paper, the automatic generative adversarial idea is combined with the design and analysis of cryptographic algorithm, and a scheme to realize componential automatic design for cryptographic algorithm is proposed by utilizing the generative adversarial method. First, a decipher is built after a cryptanalysis based on deep neural network, and the decipher experiences an automatic adversarial learning with the cryptographic algorithm, realizing an automatic design of block cipher algorithm. Then, a symmetric cryptography model of generative adversarial network is designed by analyzing the generative adversarial method. Finally, by testing different types of neural networks, the block cipher algorithm designed by the neural network is used for the encryption and decryption in the presence of third-party disturbance to the utmost extent. Meanwhile, the original symmetric cryptography model is modified on the basis of the generative adversarial method. Strengthening the attacks from third-party helps the designed cryptographic algorithm defense against different types of attacks, improving the security of cryptographic algorithm.

关 键 词：深度学习 生成对抗网络 密码算法自动设计 

分 类 号：TN91[电子电信—通信与信息系统]