Implicitly-supervised learning and its application to fuzzy pattern classifiers

The question of learning and adaptation becomes a focal point of theoretical and applied research in intelligent systems. The mechanisms of learning exhibit various facets as far as an available character of supervision is concerned as well as the quality and specificity of learning information goes. The study is devoted to the problem of learning where patterns are labelled in a heterogeneous format. The heterogeneity of labelling implies an existence of at least two specificity levels in class assignment. At the higher level of specificity, the patterns are fully labelled so that all membership values are provided. At the lower specificity level, the information about individual class memberships is replaced by its synthetic, a so-called implicit form. Quite commonly the knowledge about classes comes in a referential format. The training is guided by pairs of patterns whose similarity (or difference) degrees are specified.