Informatica

Small training sample effects common in statistical classification and artificial neural network classifier design are discussed. A review of known small sample results are presented, and peaking phenomena related to the increase in the number of features and the number of neurons is discussed.

The structure of the weight function of a random space–time autoregressive field, existing in three-dimensional space and time, is considered. The two weight coefficients calculation algorithms are proposed here.

The rules of classification of the group of N independent observations into one of k normal populations are considered. In the case when parameters are not known, Bayes estimators of the discriminant scores are suggested.

The special class of networks are presented. Based on unreachable parts of subgraphs the overlappingly decomposed networks are defined. The special decomposition scheme of those networks is applicated for shortest path problem, dynamic programming and synthetic neural nets architecture.

The checking fidelity of the systems with built-in fault detection circuits is investigated here taking into account the failure rates of functional units, also checking and transmission facilities of right and wrong information on the state of functional units. When fault detection circuits or information transmission facilities fail, then the information given by them may cause false or undetected failures in comparison with the real state of the corresponding functional unit.

Binary logic functions ‘AND’ and ‘OR’ of negations are realized by a dendritic branch with nonlinear current-voltage characteristic of membrane. The neuron with such dendrites is a complex logic system performing a great number of elementary logic operations.

Some exceptional features of the dynamically changing environment require the additional means for qualitative knowledge representation, data verification and ensurance of operativity of decision making processes. The article considers the possibilities of integration the safequard methods for impartial multicriteria: decision making with the three levels of knowledge representation distinguished in the decision support system (DSS). The approach is to represent static and dynamic aspects of the target system and to reflect them in deep knowledge representation level. The means to formalize multiple objective decision making mechanisms is proposed. The examples developed during the designing stages of the ecological evaluation system for different enterprises demonstrate the results of using extended E-nets for modeling cognitive processes leading to decisions.

In the Jasinevitchius (1991) we have introduced the new approach to the parallel space–time, computing structures (PASTICS). This paper deals with the further elaboration of the concept and synthesis of a particular architecture dedicated to the computer visions systems.