A New Approach on Neural Cryptography with Dynamic and Spy Units Using Multiple Transfer Functions and Learning Rules

摘要

There is a necessity to secure the message when the exchange of secret information is taken place among the intended users. We can generate a common secret key using neural networks and cryptography. Two neural networks which are trained on their mutual output bits are analyzed using methods of statistical physics. In the proposed TPMs, hidden layer of each output vectors are compared, then updates from hidden unit using Hebbian learning rule, left-dynamic unit using Random walk rule, right-dynamic unit using Anti-Hebbian learning rule, lower layer spy unit and upper layer spy unit with feedback mechanism. Also, we increase the effective number of keys using entropy of the weight distribution against brute-force attack. The genetic attack, geometric attack and majority attack are also explained in this study.