Informatica

There exist two principally different approaches to design the classification rule. In classical (parametric) approach one parametrizes conditional density functions of the pattern classes. In a second (nonparametric) approach one parametrizes a type of the discriminant function and minimizes an empirical classification error to find unknown coefficients of the discriminant function. There is a number of asymptotic expansions for an expected probability of misclassification of parametric classifiers. Error bounds exist for nonparametric classifiers so far. In this paper an exact analytical expression for the expected error EPN of nonparametric linear zero empirical error classifier is derived for a case when the distributions of pattern classes are spherically Gaussian. The asymptotic expansion of EPN is obtained for a case when both the number of learning patterns N and their, dimensionality p increase infinitely. The tables for exact and approximate expected errors as functions of N, dimensionality p and the distance δ between pattern classes are presented and compared with the expected error of the Fisher's linear classifier and indicate that the minimum empirical error classifier can be used even in cases where dimensionality exceeds the number of learning examples.

Weak approximation methods for initial value problem for the parabolic equation are considered. We propose some simple tests to investigate the quality of RNG used in Monte-Carlo simulations. Numerical examples are given to illustrate the application of stochastic approximation methods.

We consider a stochastic algorithm of optimization in the presented paper. We deal here with the average results of a “mixture” of the deterministics heuristics algorithm and uniform random search. We define the optimal “mixture”.

The paper describes the use of adaptive and non-periodic sampling in different fields of System Theory and Control. The review is organized in a very comprehensive way and it presents results of the last thirty years about the problem of signal applications using as main tool adaptive sampling schemes including results is the improvement of the transient behaviors. Also, related results are presented about the use of non-periodic sampling in compensation as an alternative design to the well-known frequency domain methods and about the choice of the sampling points in order to improve the transmission of measuring and/or rounding errors towards the results when studying the properties of dynamic systems such as controllability, observability and identifiability.

This paper is devoted to the consideration of the evolution of the non-migrating limited panmiction population taking into account the size, sex and age structure, pregnancy and females restoration period after delivery. The unique solvability of this model and the condition for the population to vanishe is obtained.

Recurrent neural networks of binary stochastic units with a general distribution function are studied using Markov chains theory. Sufficient conditions for ergodicity are established and under some assumptions, the stationary distribution is determined. The relation between fixed points and absorbing states is studied both theoretically and through simulations. For numerical studies the notion of almost absorbing state is introduced.

Multidimensional scaling (MDS) is well known technique for analysis of multidimensional data. The most important part of implementation of MDS is minimization of STRESS function. The convergence rate of known local minimization algorithms of STRESS function is no better than superlinear. The regularization of the minimization problem is proposed which enables the minimization of STRESS by means of the conjugate gradient algorithm with quadratic rate of convergence.