| With the rapid development of computer technology and communicationtechnology, information security has played a very important role in informationtransmission and information storage. The development of cryptography also provides astrong protection. Cryptography is a science which researches on how to transmitinformation secretly. Actually, lots of knowledge of cryptography has been extensivelyapplied in the fields of engineering, it has been a comprehensive and sophisticatedscience.Neural cryptography is a branch of cryptography research appeared nearly tenyears ago, which aims at key-change through neural network learning from each other.In2002, German scholar I.Kanter proposed the concept of neural cryptography anddemonstrated the feasibility of the neural cryptography, at the same time he put forwardseveral learning rules and verified them through a lot of theoretical analysis andstatistical experiments. After that, many researchers proposed some new methods oflearning which made neural cryptography a further development. But, throughout theseresearch results, most of them is either the dynamic analysis of the existing learningmethod or the feasibility of the proposed method. The main content of this paper is tointroduce some new methods to improve the efficiency of neural synchronization. Themain contents of this paper are listed as follows.â‘ The concepts of neural synchronization are introduced. First, this chapter givesan introduction of Tree Parity Machines which are used in neural synchronization. Thensome important parameters in the process are introduced. Finally, it introduces how toapply tree parity machines model into neural key exchange protocol.â‘¡This chapter makes a detailed introduction of the dynamics of neuralsynchronization process. First, it introduces the impact of learning rules which lead toattract steps and repulsive steps, then the random walk of overlap and synchronizationtime is analyzed.â‘¢This chapter analyzes the security of neural cryptography. First, the successprobability of an attacker which uses simple attack or main attack is analyzed. Then thesecurity of interaction which contains version spaces, mutual information and effectivekey length is introduced.â‘£Two new learning rules which improves the classic learning rules such as Hebbian and Random walk learning rules are proposed. Then, lots of simulationexperiments demonstrate the improved performance. Finally, the security of newmethods is analyzed. |
