期刊名称：International Journal of Computer Science and Network
印刷版ISSN：2277-5420
出版年度：2018
卷号：7
期号：2
页码：83-98
出版社：IJCSN publisher
摘要：This research aims to demonstrate the functionalities and prediction capabilities of different neural networks onsubstitution cipher. To acquire the best possible extent of applicability all possible plaintexts are verified. The normalized FeedForward Back Propagation Network with Feature Scaling approach results in 50% - 60% of accuracy whereas N – state SequentialMachine with Jordan Network enhanced the accuracy to 100%. This comparative study illustrates the strength of recursive backpropagationover simple back-propagation technique. The incompatibility to deal with plaintext with alphanumeric value makes NStateSequential Machine less efficient. Further, in comparison with former researches on N-State Sequential Machine; that wasdesigned for 3-bit plaintext input with alphabet range [A-H], modifications are performed, to increases the accepted input bit rangefrom 3-bit to 5-bit. This increases the input alphabet acceptance range [A-Z] including few punctuations marks- but falls short todeal with ‘alphanumeric values’. Finally, Chaotic Network is implemented and proved to be a most promising, efficient, significantand well-suited technique to deal with all input plaintexts.
关键词：Chaotic Neural Network; Cryptography; Feature Scaling; Jordan Network; Modified Caesar Cipher; N – State;Sequential Machine; Pangrams; Unity Based Normalization.
