Publication

Abstract : Data integrity is a crucial part of any secure system. Cryptographic hash functions serve as a basic building block of information security for ensuring data integrity and data origin authentication. They are used in numerous security applications such as digital signature schemes, construction of MAC and random number generation. A hash function takes an arbitrary amount of input and produces an output of fixed size. Many of the widely used cryptographic MD-5 and SHA-1 hash functions have been shown to be vulnerable to attacks. The non linear behavior of the neural network model which takes multiple inputs to produce single output makes it a perfect entrant for cryptographic hash design. The paper describes the construction of a cryptographic hash function using a multi layer Tree Parity Machine neural network. Although in our simulations we have considered 512 bit message blocks which generate 128 digit hash value, the proposed algorithm can be used flexibly to generate a hash function of arbitrary length. Simulations show that this hash function satisfies the security requirements of confusion, diffusion, and collision attack.