一种基于队列机制的神经密码学的新学习规则

New learning rule of neural cryptography based on queue mechanism

针对神经密码学中现有队列机制学习规则安全性不足的问题,提出了一种新的安全性更高的学习规则。该学习规则修改了原学习规则改变权值的幅度,使权值修改的幅度不会因为突触深度的增加而增加,从而避免了因权值修改的幅度过大使神经密码的安全性降低。最后利用概率统计实验对该学习规则在简单攻击和几何攻击下的安全性进行仿真。实验结果证明该学习规则与改进前的学习规则和经典学习规则相比,安全性均有所提高。

For the problem of inadequate security of the existed queue learning rule, this paper proposed a new security-en- hanced learning rule. The learning rule modified the magnitude of the weight change of the original learning rule so that the magnitude of the weight change didn＇ t increase with the addition of the synaptie depth, thus avoiding decreasing the security of the neural password because of the excessive weight change magnitude. Finally, the paper used the experiments of probabi- lity and statistics to simulate the learning rule under the simple attack and the geometric attack. Experimental results show that compared with the original queue learning rule and the Hebbian learning rule, the security of the new queue learning rule has an improvement.