Search academic texts

Information × Registration Number 0211U005740, 0111U007107 , R & D reports Title Development of theoretical principles and methods of realization the open trained system of the automatic recognition: methods of optimization the training samples and methods of construction weighted decision rules of classification popup.stage_title Head Volchenko Elena Vladimirovna, Registration Date 26-12-2011 Organization Hicher education establishment "Donetsk national tecnical university" popup.description2 The open adaptive trained systems of automatic recognition are an object of investigation. The algorithms of training samples optimization and methods of constructing decision rules of classification are the subject of investigation. The development of theoretical foundations and methods of implementation of open adaptive systems of automatic recognition is the aim of work. The methods for the conversion of input training samples in the weighted sample of w-objects, ways to add data to the training samples of w-objects, methods of constructing decision rules on weighted training samples of w-objects, criteria for assessment of the need to adjust the decision rules of classification on the new added data, ways of adjusting decision rules with minimal time and resource costs have been developed, theoretically grounded and experimentally studied. The number of results with scientific and practical value was obtained in a work. Consistent and grid methods for constructing reduced training samples of w-objects were proposed. A metric to assess proximity of w-objects in the feature space and methods of to determine the frequency of the systems adaptation was developed. Expansion of known effective methods of decision rules construction on weighted training samples of w-objects was performed. The usage of weighted training samples to construct recognition systems increases the efficiency of classification, solves the problem of unbounded growth of training samples and significantly reduces the time and resource costs of performing the classification. The proposed methods can extend the range of problems that decide by the construction of recognition systems. It leads to the creation of new automatic systems for various branches of knowledge. Product Description popup.authors Волченко Олена Володимирівна popup.nrat_date 2020-04-02 Close