Vol 21, No 12 (2015)

Human associative memory is a tool for negotiation of information complexity which is a result of its living. In the paper shown an architecture, properties and functions of associative memory for its using in negotiation of information complexity. A rearranging process of weight characteristics of input event images stored in columns hierarchy of hemisphere of human cortex which is fulfilled in hippocampus lamellas are shown too. A representation of human world model in cortex which are formed on base of several levels and several modalities vocabularies hierarchy is shown. Isomorphism of linguistic and multimodal human world models is represented too.

The application of genetic algorithm to finding a topology of an artificial neural network as a part of a gas-analytical device of "elettronic nose" kind is considered for odor pattern recognition. The search of the topology has been carried out via optimization to increase the recognition rate and to reduce the computing cost by genetic algorithms as a criterion for predefined possible parameters values. The optimized parameters are the number of hidden layers, the number of neurons in each layer and the training set size. The algorithm code and data processing have been realized in MATLAB software. The odor pattern recognition has been produced by feed-forward artificial neural network with Levenberg-Marquardt training method. The received results approve the efficiency of a proposed method to ANN topology finding in electronic nose devices.

Problem of clustering trajectories is pre-conditioned by a need to organize motion of objects under control. The polynomial regression method is the best approach to trajectory cluster selection, which estimates a form of general trajectory of each cluster. For a set of sufficiently heterogeneous trajectories, the defined clusters are heterogeneous also. For the inhomogeneous cluster, its polynomial regression is a too strong abstraction. To demonstrate the cluster heterogeneity (and, thus, non full clustering trajectories) a method of dimension reducing is in need. As K d-flats algorithm is a generalization of K-means algorithm where d-dimensional best fit affine sets replace centroids as the cluster prototypes, a similar modification of spectral clustering method is in use in paper. This is geometric method to obtain vector clustering model in form of affine space union. Vector set similarity is estimated by multi-dimensional affinity tensor, for which matrix representation is in use. Next it is analyzed by spectral method. As example, clustering results of airplane fly trajectories in an airport space is under study. Defining the cluster central trajectory with affinity estimation by cosine measure in use gives an opportunity to determine the cluster outlying trajectories in potential, which represent fly trajectories located in a risk zone.

Description of the system consisting of a large set of tests and a judge program is proposed. The system is used for performance evaluation of arbitrary algorithms of protection of logic and arithmetic circuits without feedback. While developing the system, experience in preparing tasks for sports programming competitions was used. The paper presents the details of the closed system of tests development and of the scoring methodology. Details of the judge program work are revealed. The developed test system is freely available online from the SPOJ system for all reliability professionals. The system ranks the algorithms and arranges them in a list according to their total score. The more points the algorithm received, the more reliable circuits based on the initial version it generated.

The paper presents the results of a study the performance of different Residue Number System moduli sets for digital signal processing based on the Finite Impulse Response filters. It is shown that the best filtration rate provide by moduli sets {2ⁿ — 1,2ⁿ, 2ⁿ + 1,2ⁿ - $2^{\frac{n+1}{2}}$ + 1,2ⁿ+ $2^{\frac{n+1}{2}}$ + 1} and {2ⁿ — 1,2ⁿ, 2ⁿ + 1,2^{n + 1} — 1}. The difference in performance between the best and the worst moduli sets up to 93 %. The results can be used to improve the performance of the digital signal processing devices used Residue Number System and applied to the processing of images, video, speech and other applications.

We introduce vulnerability threats classification as a basic informational security element (modeling object) in borders of which threats are divided to technological and implementation ones (which create threats of technological vulnerabilities). We show that implementation vulnerability threats must be researched as vulnerability threats modeling object (for attack modeling). While this modeling can be done without any expert assessments, i.e. just by solely using existing statistics about security parameters of vulnerabilities threats. We substantiate Markov models with discrete states and continuous time (widely used in reliability theory for modeling of systems failures and recoveries) usage correctness to model attack threats. However, we show the necessity of inclusion into vulnerability threat model of "birth and death" scheme elements (because of same time similar vulnerability possibility which create the same threat of attack). This produces fundamentally differences from the modeling problems of reliability theory. We suggest vulnerability threat serial reservation scheme as an attack threat interpretation (which defines informational system attack threat modeling approach with use of built queuing models). We show an example of vulnerability and attack threats quantitative relevance assessment (produced by single implementation vulnerability threat) illustarting possibilities of the proposed modeling approach.

The complex of models needed for the development of current and future neurochips and neuromorphic computers based on them is presented. The complex of models includes models related to the development of neural network algorithms for solving problems, process models, associated with the construction of the basic elements of the circuitry, and circuit design model concepts realizing neurochips and functional elements.

Due to complexity of current and future neurochips and neuromorphic computers, it is necessary, before their creation, to develop their complex models for testing their basic features and functions. The combination in modern and future neurochips and neuromorphic computers such different technologies as memristors, analog CMOS, digital CMOS, optical, various techniques of development of neuroboards, neroblocks, neuroracks and superneurocomputers, makes it necessary to develop and use such complex models, CAD systems and elements libraries.

This article describes these models and give an explanation of the objective necessity of the implementation such software models on personal computers and GPU-supercomputers.