一种改进的基于树形奇偶机的神经网络同步方案

An Improved Scheme for Neural Synchronization Based on Tree Parity Machine

神经网络通过相互学习达到完全同步来构建密码协议,已经成为当今密码学的重要研究方向。针对神经密码同步过程中通信次数过多的问题,本文利用神经密码应用中的树形奇偶机,在综述神经密码协议研究的基础上,提出对神经网络单元初始权值的选取进行改进的解决方案。仿真实验结果表明,在保证安全性的情况下,本文方案大大加快了同步速度。

Neural network can fully synchronize by learning from each other, because this effect neural synchronization can be used to construct a cryptographic key-exchange protocol, which has become an important research direction in current cryptography. To solve the problem of too many times in neural cryptography synchronization, an improved scheme was proposed by employing Tree Parity Machine （TPM）. On the basis of neural cryptographic protocols, the range of initial weights of neural units was appropriately reduced. After the analysis, the simulation results show that in the case of ensuring security, synchronization efficiency is greatly improved by applying new improvement scheme.