Ostroverkhyi O. Neural network processing of multidimensional signals

Object of the research is the processes of adaptive processing of multidimensional information in conditions of antecedent and current uncertainty regarding the properties of the objects under consideration and applied noise based on Artificial Neural Networks (ANN). Goal is the development of ANN models and methods for improving the efficiency of real-time multidimensional data processing by decreasing the hardware and computational costs while keeping or improving the functional characteristics. Research methods are based on the theory of computational intellect, particularly on the theory of ANNs, and the theory of optimization. Theoretical and practical results in aggregate are the solution of a topical scientific and practical problem related to the improvement of the efficiency of real-time adaptive multidimensional data processing. Scientific novelty: 1) first introduced method for automatic creation of the input space adaptive sampling schemes for CMAC ANN and its derivatives, which is based on the analysis of learning sequences entropy by calculating the Grey relational coefficient; 2) gained further development the modular CMAC ANN models for multidimensional data processing which are the linear or hierarchical combinations of the low-order CMAC ANNs, in particular by minimizing their structure and input space self-arrangement; 3) improved the training procedures of the feed-forward finite impulse response dynamic neural networks by rearranging the learning sums in the temporal back-propagation algorithm. The scientific and practical results are introduced at Scientific Research Complex "Accelerator" of National Scientific Center "Kharkiv Physical Technical Institute" and at Computers department of Kharkiv National University of Radio Electronics. Area of application includes the organizations dealing with the problems of developing the intellectual systems for adaptive data processing and in training courses for preparation of the specialists in the area of intellectual information processing systems.