Cryptosystem for textual message by using malleable neural network with secret dimensions

Abstract

Most of the current cryptography is based on modulo computing with predefined public and private keys. The robustness of this approach is controlled by the size of the prime divider. The encryption and decryption processes of this modulo computing are rather fast and rather robust to any attack. However, it is still possible to crack the encrypted message by using meta-heuristic algorithm, very high performance parallel machines, and also quantum computing concept. This thesis takes another approach in order to enhance the robustness of encrypted messages by deploying the structure of a feedforward neural network with augmented secret dimensions concurred by both sender and receiver. The network structure can be adjusted to make it difficult to guess. The proposed concept differs from weight synchronization and weight representation of the encrypted message. The text message is the main concern of this study. However, the proposed concept can be extended to cope with other forms of input messages. The experimental results showed the robustness of encrypted messages against attacks by using cryptanalysis. The space and time complexities of the network and process are discussed.