KOUDELKA, V. Pravděpodobnostní neuronové sítě pro speciální úlohy v elektromagnetismu [online]. Brno: Vysoké učení technické v Brně. Fakulta elektrotechniky a komunikačních technologií. 2014.
Dissertation thesis written by Vlastimil Koudelka was focused on the development of special neural networks to be used for the solution of selected problems from the area of applied electromagnetics. The research topic was inspired by the FP7 project High Intensity Radiated Field – Synthetic Environment (HIRF-SE). Here, special post-processing tools were required to be developed for probabilistic modeling of electromagnetic fields and for the simplification of highly complicated structures (cable bundles, dominantly). Vlastimil developed corresponding networks and integrated them to the platform for the virtual EMC certification of small airplanes. Outputs of the research were published in IET Microwaves, Antennas and Propagation, Radioengineering, and several proceedings of European conferences. At present, a paper Evidence-based selection criterion for probabilistic and general regression neural networks is in a review process at Elsevier’s journal Neural Networks. During his studies, Vlastimil spent two months in the laboratory of Prof. Carlos del Rio Bocio (University of Navarra, Spain) exploiting artificial neural networks for microwave imaging. He also attended the European School of Antennas and International Travelling Summer School on Microwaves and Lightwaves. An original contribution of the thesis written by Vlastimil can be found in two directions: • Boltzman machine. An analogy in between the structure of the neural network and the structure of the problem to be simplified was formulated. When training the neural network (and optimizing its structure), the problem to be simplified is solved at the same time. • Probabilistic neural networks. An accurate and computationally efficient way of estimating the probability density function in probabilistic neural networks was developed. Vlastimil showed that the crucial problem of the Bayesian approach can be solved by appropriate combinatorial and likelihood approximations. Vlastimil significantly contributed to the European HIRF-SE project, and to the national project Millimeter-wave electromagnetic structures for biomedical research. Outputs of Vlastimil’s work are highly appreciated, and his thesis is recommended to be defended.
viz posudek ve formátu pdf
viz posudek ve formátu pdf
eVSKP id 72666