. "2010-01-01+01:00"^^ . " retrieval" . " decisions" . . "Bregman" . "Projekt b\u011Bhem sv\u00E9ho trv\u00E1n\u00ED zm\u011Bnil hlavn\u00EDho \u0159e\u0161itele a \u010D\u00E1ste\u010Dn\u011B i sv\u00E9 c\u00EDle. C\u00EDle projektu lze hodnotit jako spln\u011Bn\u00E9. V\u00FDsledky byly publikov\u00E1ny v mezin\u00E1rodn\u00EDch \u010Dasopisech a sborn\u00EDc\u00EDch mezin\u00E1rodn\u00EDch konferenc\u00ED."@cs . " machine" . "1"^^ . "The principal investigator changed during the project. The goals of this project were fulfilled. The results were published in international journals and proceedings of international conferences."@en . . " distributions" . "Bregman; distances; divergences; of; distributions; information; retrieval; optimal; decisions; machine; learning"@en . "http://www.isvav.cz/projectDetail.do?rowId=GAP202/10/0618"^^ . "0"^^ . "2012-03-30+02:00"^^ . . "0"^^ . . . . " divergences" . . . . . " distances" . "Bregman distances, divergences, and optimal algorithms for information retrieval, decisions and learning"@en . . . "5"^^ . " optimal" . "5"^^ . . "GAP202/10/0618" . . "Bregmanovy vzd\u00E1lenosti, divergence a optim\u00E1ln\u00ED algoritmy pro hled\u00E1n\u00ED informac\u00ED, rozhodov\u00E1n\u00ED a u\u010Den\u00ED" . "2014-01-30+01:00"^^ . " of" . "V\u00FDzkum je zam\u011B\u0159en na kriteria shody zn\u00E1m\u00E1 jako Bregmanova vzd\u00E1lenost a f-divergence a jejich aplikace ve strojov\u00E9m zpracov\u00E1n\u00ED informace. Ob\u011B kriteria jsou v praxi aplikov\u00E1na na diskr\u00E9tn\u00ED distribuce z\u00EDskan\u00E9\u00A0\u010Dasto pomoc\u00ED razantn\u00EDho kvantov\u00E1n\u00ED v prostorech slo\u017Eit\u00FDch p\u0159\u00EDznak\u016F datab\u00E1zov\u00FDch objekt\u016F, jako nap\u0159\u00EDklad spektr\u00E1ln\u00EDch hustot nebo \u010Dasov\u011B prom\u011Bnn\u00FDch charakteristik. Snaha zv\u00FD\u0161it optim\u00E1lnost pou\u017E\u00EDvan\u00FDch algoritm\u016F vytv\u00E1\u0159\u00ED tlak na omezov\u00E1n\u00ED informace ztracen\u00E9 p\u0159i reprezentaci objekt\u016F. To vy\u017Eaduje aplikovat zm\u00EDn\u011Bn\u00E1 kriteria na distribuce mnohem slo\u017Eit\u011Bj\u0161\u00ED povahy, co\u017E v prv\u00E9 \u0159ad\u011B vy\u017Eaduje roz\u0161\u00ED\u0159en\u00ED jejich definic na matematicky zna\u010Dn\u011B abstraktn\u00ED prostory m\u011B\u0159iteln\u00FDch hustot ne nutn\u011B normovan\u00FDch k jedni\u010Dce. Tento krok a prozkoum\u00E1n\u00ED vz\u00E1jemn\u00FDch vztah\u016F takto roz\u0161\u00ED\u0159en\u00FDch obou kriteri\u00ED a odvozen\u00ED jejich vlastnost\u00ED vyu\u017Eiteln\u00FDch v procesu jejich aplikace tvo\u0159\u00ED hlavn\u00ED p\u0159edm\u011Bt navrhovan\u00E9ho projektu. D\u016Fraz je kladen tak\u00E9 na metrick\u00E9 vlastnosti obou charakteristik, na jejich vztahy ke klasick\u00FDm informa\u010Dn\u011B-teoretick\u00FDm a statistick\u00FDm pojm\u016Fm a demonstraci zlep\u0161en\u00FDch v\u00FDsledk\u016F p\u0159i hled\u00E1n\u00ED informac\u00ED v datab\u00E1z\u00EDch, p\u0159i rozhodovac\u00EDch procesech ve zpracov\u00E1n\u00ED obrazu a \u0159e\u010Di a ve strojov\u00E9m u\u010Den\u00ED." . . " information" . . "The research deals with\u00A0the similarity criteria called Bregman distances and f-divergences\u00A0 and their applications in machine information processing. These criteria are so far applied in the praxis to discrete distributions, obtained\u00A0often by rough quantizations in the feature spaces of database objects of complex functional nature such as spectral densities or various time-varying characteristics. The need to increase the optimality of used algorithms creates pressure on extension of application of these criteria to more complex distributions. This means in the first place to extend their definitions to more abstract spaces and to investigate the mathematical properties relevant for applications. The objectives of this project is precisely this step, namely the extension\u00A0of both these criteria to arbitrary measurable densities not necessarily normalized to 1 and summarization of their mutual relations and properties for references in the process of application. A particular emphasis is on metric properties, relations to the classical information-theoretical and statistical concepts and demonstration of improved applicability in information retrieval, decision processes related to image and speech processing and machine learning."@en . . "2012-12-31+01:00"^^ . . .