生成式对抗网络的通信网络安全技术

Communication Network Security Technology Based on Generative Adversarial Networks

为了更好地解决通信网络中存在的恶意攻击,保护用户数据安全,通过生成式对抗网络的生成模型和判别模型相互博弈不断优化,构造成最优判别器,可以对数据攻击进行检测。主要介绍生成式对抗网络(GANs)和Wasserstein生成式对抗网络的区别、模型及算法,通过研究GAN与WGAN梯度消失问题,实验证实WGAN可以有效地解决网络收敛性差、模型自由不可控、训练不稳定等问题,具有更好的性能。

In order to better solve the current malicious attacks in the communication network and protect user data security,the optimal discriminator is constructed via mutual gaming and optimization between generative and discriminant models of generative adversarial networks(GAN)and thus it can be used to detect data attacks.This article mainly introduces the differences,models and algorithms of both GANs and Wasserstein GANs(WGAN)via investigating the corresponding issue of gradient disappearance.The experiments prove that WGAN can effectively solve problems such as poor network convergence,uncontrollable model,unstable training and yield a better performance.