Abstract. The paper presents a general approach to the evaluation of the complexity of classes of algorithms, the so-called pVCD-method. To develop this method, all the examined families of models of empiric generalization were limited to classes implementable on computers and wider, by examining their partly recursive presentations. Within the framework of the algorithmic approach, the concept of Kolmogorov’ complexity of classes of algorithms of the recognition of properties or extraction of regularities is proposed. Based on this concept, a method is proposed to evaluate the nonrandomness of the extraction of empirical regularities. Tabl.: 2. Refs: 6 titles.

Keywords: machine training, recursive functions, VC-dimension, Kolmogorov complexity.