互学习神经网络集合的同步

Synchronization of Interacting Neural Networks

互学习的神经网络特别是树状奇偶模型的神经网络因能通过一定量的信息交换达到同步而被广泛地应用在密码学等领域。为了扩展该同步模式的用途,提出一种由多个树状奇偶机组成的神经网络集合的同步方式,分布式同步方式(两两成对同步),并在此基础上讨论分布式同步与向中心学习的同步方式相关性能及比较结果。实验表明,分布式同步在时间复杂度和系统复杂程度上都具有一定的优势,是一种较好的多神经元同步模式,为密钥分发提供了一种新的理论与应用研究方向。

Several scenarios of interacting neural networks especially Tree Parity Machine are widely used in many fields such as cryptography. This kind of neural networks has some good properties such as they can reach a same state by transmitting some information limited. In this paper, the synchronization patterns of Tree Parity Machines are discussed, named learn-neighbor pattern and distributed pattern. Based on the schemes of the two patterns, the advantages of the patterns are investigated, as well as the differences and relations. The experimental resuits show that the distributed pattern is better than the learn-neighbor pattern on the time complexity and the system architecture is simple.