摘要
机器学习的不断发展对现代密码体制造成的威胁不断增大,如何将神经网络应用到密码学上的研究日益深入,而生成对抗网络的对抗学习机制与密码学中的对抗性质相符,因此,研究了对抗网络与密钥协商相结合的问题。作为初步的概念验证,直接采用神经网络代替通信双方和敌手,利用对抗学习机制为核心思想设计对抗网络下的密钥协商模型(key agreement based on adversarial network,KG-AN),并进行了密钥长度为16 bit和64 bit的训练。实验结果显示,通信双方的协商密钥误差分别在1.5%和2%左右,敌手的破译误差分别保持在95%和91%左右,初步实现了对抗网络下的密钥协商功能,验证了对抗网络应用到密钥协商的可行性。
With the continuous development of machine learning,the threat to modern cryptography is constantly increasing,and the research on neural network applied to cryptography is also deepening,and adversarial learning mechanism in generative adversarial networks is consistent with the adversarial nature in cryptography.This paper studied how to combine adversa-rial network and key agreement.As a preliminary proof of concept,this paper directly used the neural network instead of both sides and adversary of communication,and used the adversarial learning mechanism to design the key agreement based on adversarial network(KG-AN),and the key length was 16 bit and 64 bit of training.The experimental results show that both sides’agreement key error of the communication is 1.5%and 2%respectively,the decoding error of the adversary remains at around 95%and 91%respectively,which preliminarily realizes the key agreement under the adversarial network,and demonstrates the feasibility of applying the adversarial network to the key agreement.
作者
吴少乾
李西明
王璇
蔡河鑫
Wu Shaoqian;Li Ximing;Wang Xuan;Cai Hexin(College of Mathematics&Informatics,South China Agricultural University,Guangzhou 510642,China)
出处
《计算机应用研究》
CSCD
北大核心
2021年第5期1539-1543,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61872152,61872409)
2018年广东省农业厅省级乡村振兴战略专项资助项目(粤农计〔2018〕54号)
广东省基础与应用基础重大项目(2019B030302008,2020A1515010751)
广州市科技计划资助项目(201902010081)。
关键词
对抗网络
密钥协商
对抗学习机制
反卷积网络
adversarial network
key agreement
adversarial learning mechanism
deconvolution network
