We picked Evolutionary Computation as our basic optimization technique. Main reason for this choice — properly configured evolutionary algorithm greatly outperforms all other known to us methods of optimization both in the complexity of successfully resolvable tasks and in stability to singularities of the search space. We have developed an object-oriented version of Genetic Algorithm, performed theoretic and experimental research of its performance, stated efficiency conditions (reference to Koroteyev, V.A. (1986) Methods and Algorithms for Control System Items Identification. PhD Thesis, Kiev Institute of Cybernetics).

We named the method Evolutionary Search. In addition to the optimization of the model's structure — our main application of the method, it possesses its own value and has shown its effectiveness in the tasks of scheduling, resources distribution, etc.

For Data Mining we have developed Evolutionary Regression — an algorithm that builds polynomial approximations from data. The algorithm addresses the main problem of multivariable empirical dependency modeling — search for an optimal structure of the approximation function. Evolutionary Regression applies Evolutionary Search to the search of optimal consist of the approximation polynomial. We have performed theoretic and experimental examination of the method: Evolutionary Regression was successfully applied to dependencies recognition from different practical areas.

An algorithm for building classification models is one of our most recent developments. We call it Evolutionary Classification. Most important features that we address during the design of this method were reliability and effectiveness for practical classification tasks. Such tasks usually have high number of dimensions of factors of doubtful value. Evolutionary Regression applies the power of Evolutionary Search both to select the most valuable factors and to optimize the composition of the supporting clusters. Our experiments conducted both on empirical and artificially generated data proved efficiency of the algorithm.

Last and most important direction of our research and development activities is building theory and implementation of a system capable to collect, organize and to place at reuse knowledge obtained from experience. Today we have only a very general idea on the organization of such a system and this is exactly the goal to which we are going to direct our research activity.