Abstract—Recently, there are some sorts of attacks that have been proven to be effective on the Advanced Encryption Standard AES. Thus the paper attempts to do some modification to stand against such sort of attacks by employing nonlinear Neural Network NN in the design and implementation of the AES. In the design of the NN performs encryption and decryption processes using of symmetric key cipher. The key used in both encryption and decryption processes is the initial weights for neural network and then train to its final weight with a fast and low cost algorithm, such as Levenberg – Marquardt Algorithm. The final weights of neural network represent the final key that can be used for encryption and decryption processes. The target from the network has been selected to match the output of the AES that have an efficient and recommended security. The proposed NN design has been modeled and computer simulated. Simulation results show the closeness of the results achieved by the proposed NN-based AES cryptosystem with that of the normal AES.

Index Terms—AES, Neural Networks, Symmetric Block Cipher, Levenberg – Marquardt Algorithm.

S. Y. Ameen, Professor with the Gulf University, Kingdom of Bahrain, Dean of College of Engineering (phone: 973-39304338; fax: 973-17622230; e-mail: prof-siddeeqr@ieee.org).
A. H. Mahdi., He is now with the Department of Information Engineering, Baghdad University, Bagdad, Iraq (e-mail: aliinfo1980@yahoo.com).

Cite: Siddeeq. Y. Ameen and Ali H. Mahdi, "AES Cryptosystem Development Using Neural Networks," International Journal of Computer and Electrical Engineering vol. 3, no. 2, pp. 309-314, 2011.