@@ -12,14 +12,14 @@
\contentsline {subsection}{\numberline {2.2}Met\IeC {\' o}da hlavn\IeC {\' y}ch komponentov pomocou algoritmu GHA a APEX, architekt\IeC {\' u}ra modelu, vz\IeC {\v t}ah pre adapt\IeC {\' a}ciu v\IeC {\' a}h, pojem vlastn\IeC {\' y}ch vektorov a vlastn\IeC {\' y}ch \IeC {\v c}\IeC {\' \i }sel, redukcia dimenzie, aplik\IeC {\' a}cia na kompresiu obrazu.}{12}{subsection.2.2}
\contentsline {subsection}{\numberline {2.3}U\IeC {\v c}enie so s\IeC {\' u}\IeC {\v t}a\IeC {\v z}en\IeC {\' \i }m (typu \IeC {\textquotedblleft }winner-take-all\IeC {\textquotedblright }), nev\IeC {\' y}hody. Neurobiologick\IeC {\' a} motiv\IeC {\' a}cia algoritmu SOM, later\IeC {\' a}lna interakcia a jej n\IeC {\' a}hrada v SOM, sumariz\IeC {\' a}cia algoritmu, vo\IeC {\v l}ba parametrov modelu.}{13}{subsection.2.3}
\contentsline {subsection}{\numberline {2.4}SOM: vektorov\IeC {\' a} kvantiz\IeC {\' a}cia, topografick\IeC {\' e} zobrazenie pr\IeC {\' \i }znakov, algoritmus SOM, parametre, redukcia dimenzie, magnifika\IeC {\v c}n\IeC {\' a} vlastnos\IeC {\v t}, pr\IeC {\' \i }klad pou\IeC {\v z}itia.}{15}{subsection.2.4}
\contentsline {subsection}{\numberline {2.5}Hybridn\IeC {\' e} modely NS, RBF model: aktiva\IeC {\v c}n\IeC {\' e} vzorce, b\IeC {\' a}zov\IeC {\' e} funkcie, pr\IeC {\' \i }znakov\IeC {\' y} priestor, probl\IeC {\' e}m interpol\IeC {\' a}cie, tr\IeC {\' e}novanie modelu, aproxima\IeC {\v c}n\IeC {\' e} vlastnosti RBF siete.}{17 }{subsection.2.5}
\contentsline {section}{\numberline {3}Rekurentn\IeC {\' e} a pam\IeC {\" a}\IeC {\v t}ov\IeC {\' e} modely. Ot\IeC {\' a}zky 13 a\IeC {\v z} 18.}{19 }{section.3}
\contentsline {subsection}{\numberline {3.1}NS na spracovanie sekven\IeC {\v c}n\IeC {\' y}ch d\IeC {\' a}t: reprezent\IeC {\' a}cia \IeC {\v c}asu, typy \IeC {\' u}loh pre rekurentn\IeC {\' e} NS. Modely s \IeC {\v c}asov\IeC {\' y}m oknom do minulosti, v\IeC {\' y}hody a nedostatky, pr\IeC {\' \i }klad pou\IeC {\v z}itia.}{19 }{subsection.3.1}
\contentsline {subsection}{\numberline {2.5}Hybridn\IeC {\' e} modely NS, RBF model: aktiva\IeC {\v c}n\IeC {\' e} vzorce, b\IeC {\' a}zov\IeC {\' e} funkcie, pr\IeC {\' \i }znakov\IeC {\' y} priestor, probl\IeC {\' e}m interpol\IeC {\' a}cie, tr\IeC {\' e}novanie modelu, aproxima\IeC {\v c}n\IeC {\' e} vlastnosti RBF siete.}{16 }{subsection.2.5}
\contentsline {section}{\numberline {3}Rekurentn\IeC {\' e} a pam\IeC {\" a}\IeC {\v t}ov\IeC {\' e} modely. Ot\IeC {\' a}zky 13 a\IeC {\v z} 18.}{18 }{section.3}
\contentsline {subsection}{\numberline {3.1}NS na spracovanie sekven\IeC {\v c}n\IeC {\' y}ch d\IeC {\' a}t: reprezent\IeC {\' a}cia \IeC {\v c}asu, typy \IeC {\' u}loh pre rekurentn\IeC {\' e} NS. Modely s \IeC {\v c}asov\IeC {\' y}m oknom do minulosti, v\IeC {\' y}hody a nedostatky, pr\IeC {\' \i }klad pou\IeC {\v z}itia.}{18 }{subsection.3.1}
\contentsline {subsubsection}{\numberline {3.1.1}Nerekurentn\IeC {\' e} modely}{19}{subsubsection.3.1.1}
\contentsline {subsection}{\numberline {3.2}Rekurentn\IeC {\' e} NS: princ\IeC {\' \i }p tr\IeC {\' e}novania pomocou algoritmu BPTT a RTRL. Pr\IeC {\' \i }klad pou\IeC {\v z}itia.}{20 }{subsection.3.2}
\contentsline {subsubsection}{\numberline {3.2.1}Back-propagation through time - BPTT}{21 }{subsubsection.3.2.1}
\contentsline {subsubsection}{\numberline {3.2.2}Real-time recurrent learning - RTRL}{22 }{subsubsection.3.2.2}
\contentsline {subsection}{\numberline {3.3}Elmanova sie\IeC {\v t}: intern\IeC {\' e} reprezent\IeC {\' a}cie pri symbolovej dynamike, Markovovsk\IeC {\' e} spr\IeC {\' a}vanie, architektur\IeC {\' a}lna predispoz\IeC {\' \i }cia. Model rekurz\IeC {\' \i }vnej SOM (RecSOM).}{22 }{subsection.3.3}
\contentsline {subsection}{\numberline {3.2}Rekurentn\IeC {\' e} NS: princ\IeC {\' \i }p tr\IeC {\' e}novania pomocou algoritmu BPTT a RTRL. Pr\IeC {\' \i }klad pou\IeC {\v z}itia.}{19 }{subsection.3.2}
\contentsline {subsubsection}{\numberline {3.2.1}Back-propagation through time - BPTT}{20 }{subsubsection.3.2.1}
\contentsline {subsubsection}{\numberline {3.2.2}Real-time recurrent learning - RTRL}{21 }{subsubsection.3.2.2}
\contentsline {subsection}{\numberline {3.3}Elmanova sie\IeC {\v t}: intern\IeC {\' e} reprezent\IeC {\' a}cie pri symbolovej dynamike, Markovovsk\IeC {\' e} spr\IeC {\' a}vanie, architektur\IeC {\' a}lna predispoz\IeC {\' \i }cia. Model rekurz\IeC {\' \i }vnej SOM (RecSOM).}{21 }{subsection.3.3}
\contentsline {subsection}{\numberline {3.4}Sie\IeC {\v t} s echo stavmi (ESN): architekt\IeC {\' u}ra, inicializ\IeC {\' a}cia, tr\IeC {\' e}novanie modelu, vplyv parametrov na vlastnosti rezervo\IeC {\' a}ra, echo vlastnos\IeC {\v t}, pam\IeC {\" a}\IeC {\v t}ov\IeC {\' a} kapacita.}{22}{subsection.3.4}
\contentsline {subsection}{\numberline {3.5}Hopfieldov model NS: deterministick\IeC {\' a} dynamika, energia syst\IeC {\' e}mu, relax\IeC {\' a}cia, typy atraktorov, autoasociat\IeC {\' \i }vna pam\IeC {\" a}\IeC {\v t} \IeC {\textendash } nastavenie v\IeC {\' a}h, princ\IeC {\' \i }p v\IeC {\' y}po\IeC {\v c}tu kapacity pam\IeC {\" a}te.}{23}{subsection.3.5}
\contentsline {subsection}{\numberline {3.6}Neline\IeC {\' a}rne dynamick\IeC {\' e} syst\IeC {\' e}my: stavov\IeC {\' y} portr\IeC {\' e}t, dynamika, typy atraktorov. Hopfieldov model NS: stochastick\IeC {\' a} dynamika, parameter inverznej teploty, princ\IeC {\' \i }p odstr\IeC {\' a}nenia falo\IeC {\v s}n\IeC {\' y}ch atraktorov.}{25 }{subsection.3.6}
\contentsline {subsection}{\numberline {3.6}Neline\IeC {\' a}rne dynamick\IeC {\' e} syst\IeC {\' e}my: stavov\IeC {\' y} portr\IeC {\' e}t, dynamika, typy atraktorov. Hopfieldov model NS: stochastick\IeC {\' a} dynamika, parameter inverznej teploty, princ\IeC {\' \i }p odstr\IeC {\' a}nenia falo\IeC {\v s}n\IeC {\' y}ch atraktorov.}{24 }{subsection.3.6}