"24"^^ . "neural networks; probabilistic approach; distribution mixtures; pattern recognition; textures"@en . . "24"^^ . . . "1"^^ . "0"^^ . . "Rekurentn\u00ED pravd\u011Bpodobnostn\u00ED neuronov\u00E9 s\u00EDt\u011B"@cs . "2010-12-31+01:00"^^ . . . . "\"The main theoretical motivation of the project is to propose a new statistically justified approach to recurrent neural networks. The proposed research is largely based on the original results of our previous academic project on Probabilistic Neural Networks (PNN) closed in 2000 with \"\"excellent results\"\" and on the related papers published in the last five years. The basic idea of the probabilistic approach to recurrent neural networks is to introduce feedback into the standard feed-forward architecture of PNN. We shall consider especially the principle of iterative inference mechanism originally proposed for probabilistic expert systems. The potential benefit of recurrent solutions will be verified by comparing the standard feed-forward architecture with the recurrent PNN in application to pattern recognition and texture evaluation. PNN proved to be a powerful tool to solve practical problems and simultaneously, unlike standard approaches, they are biologically interpretable in many\""@en . "Recurrent Probabilistic Neural Networks"@en . "916"^^ . . "\"Hlavn\u00ED teoretickou motivac\u00ED projektu je n\u00E1vrh nov\u00E9ho statisticky zd\u016Fvodn\u011Bn\u00E9ho p\u0159\u00EDstupu k\u00A0rekurentn\u00EDm neuronov\u00FDm s\u00EDt\u00EDm. Navrhovan\u00FD v\u00FDzkum v\u00A0\u0161irok\u00FDch mez\u00EDch vych\u00E1z\u00ED z\u00A0p\u016Fvodn\u00EDch v\u00FDsledk\u016F na\u0161eho p\u0159edchoz\u00EDho akademick\u00E9ho projektu Pravd\u011Bpodobnostn\u00ED neuronov\u00E9 s\u00EDt\u011B (PNS), kter\u00FD byl ukon\u010Den v\u00A0roce 2000 s \"\"vynikaj\u00EDc\u00EDmi v\u00FDsledky\"\" a z\u00A0n\u011Bkolika navazuj\u00EDc\u00EDch prac\u00ED publikovan\u00FDch v\u00A0posledn\u00EDch p\u011Bti letech. Z\u00E1kladn\u00ED idea navrhovan\u00E9ho pravd\u011Bpodobnostn\u00EDho p\u0159\u00EDstupu k\u00A0rekurentn\u00EDm neuronov\u00FDm s\u00EDt\u00EDm p\u0159edpokl\u00E1d\u00E1 vytvo\u0159en\u00ED zp\u011Btn\u00E9 vazby ve standardn\u00ED vzestupn\u00E9 architektu\u0159e PNS. Jednu z\u00A0mo\u017Enost\u00ED nab\u00EDz\u00ED vyu\u017Eit\u00ED iterativn\u00EDho inferen\u010Dn\u00EDho mechanismu p\u016Fvodn\u011B navr\u017Een\u00E9ho v\u00A0pravd\u011Bpodobnostn\u00EDm expertn\u00EDm syst\u00E9mu. Potenci\u00E1ln\u00ED p\u0159\u00EDnos rekurentn\u00ED architektury bude ov\u011B\u0159ov\u00E1n ve srovn\u00E1n\u00ED sestandardn\u00ED vzestupnou PNS p\u0159i aplikaci v\u00A0rozpozn\u00E1v\u00E1n\u00ED a p\u0159i vyhodnocov\u00E1n\u00ED textur. PNS jsou prakticky pou\u017Eiteln\u00E9 a sou\u010Dasn\u011B p\u0159ipou\u0161t\u011Bj\u00ED biologickou interpretaci a\u017E na \u00FArovni funk\u010Dn\u00EDch vlastnost\u00ED neuronu. Neuromorfn\u00ED vlastnosti jsou v\u00FDznamnou p\u0159ednost\u00ED PNS\""@cs . . . . "916"^^ . . "2007-01-01+01:00"^^ . . "0"^^ .