Volume 4, Issue 2

We give a survey of different techniques for hiding data in several protocols from the TCP/IP protocol stack. Techniques are organized according to affected layer and protocol. For most of the covert channels its data bandwidth is given.

Cryptographic hash functions are important cryptographic techniques and are used widely in many cryptographic applications and protocols. All the MD4 design based hash functions such as MD5, SHA-1, RIPEMD-160 and FORK-256 are built on Merkle-Damgård iterative method. Recent differential and generic attacks against these popular hash functions have shown weaknesses of both specific hash functions and their underlying Merkle-Damgård construction. In this paper we propose a hash function follows design principle of NewFORK-256 and based on HAIFA construction. Its compression function takes three inputs and generates a single output of 256-bit length. An extra input to a compression function is a 64-bit counter (number of bits hashed so far). HAIFA construction shows strong resistance against major generic and other cryptanalytic attacks. The security of proposed hash function against generic attacks, differential attack, birthday attack and statistical attack was analyzed in detail. It is shown that the proposed hash function has high sensitivity to an input message and is secure against different cryptanalytic attacks.

In this article we investigate some approaches to the artificial neural networks training with use of hybrid algorithms. Algorithms which are based on the back propagation algorithm and the ant colony algorithm are considered in detail. The article describes the application of the artificial neural network with the authors’ hybrid training algorithm. The preliminary studies have shown that the algorithm improves the efficiency of the problems on standard test databases. The application of the algorithm for practical problems solution in the field of medicine, namely the definition of danger level determination of tuberculosis carriers is described. It was shown that the accuracy of the hybrid algorithm is up to 22% higher than of the classical one.