Processor
zadaniy
Neural set
zaprosov
Predobrabotchik
Planirovshchik
Prikladnaya
Client
zadaniy
program
Database primerov
neuroservice
Executor
Base model
"Neuroservice client". "Client neuroservice" is implemented in the form of ActiveX
objects and can be caused by any means of supporting technology
identifikatora zaprashivaemoy protsedurÿ, ee parameters, imeni
s nim osushchestvlyaetsya iz prikladnyx programm cherez obrashchenie k metodam
obrabotkoy zaprosa podrazumevaetsya vÿdeleniya iz XML-dokumenta
“Base primerov”
Module organizes storage and access to prime. Takje xranyatsya
XML-documents and sends "Processor queries". Answer ot
ActiveX.
In most cases, all experiments were performed using MS Excel. "Client
neuroservice" translates the call to action programs in
users and passwords.
prikladnoy programs skryvayutsya slozhnosti standard XML. Pri
svedeniya o proektax. Primers are grouped by projects. Module
"Query processor" also comes in the form of XML-document, but from
“Query Processor”
Module serves for primary query processing, identification, authentication and
application of other modules. Under the pervichnoy
Prikladnaya programma i “Klient neyroservera”
Neyroserver ne imeet polzovatelskogo interface.
Vzaimodeystvie
neobxodimosti mojno obmenivatsya soobshcheniyami s “Obrabotchikom
zaprosov ”napryamuyu. Formaty XML-dokumentov privedeny v prilojenii B.
112
Rice. 16. Modules verxnego urovnya dekompozitsii i napravlenie potokov dannyx. The
arrows indicate the direction of the exchange soobshcheniyami between modules.
Machine Translated by Google
"Ispolnitelyam zadaniy".
i mojet soderjat in sebe razlichnye svedeniya. For example, datu i istochnik
otvet INS on osnovanii predÿyavlennyx vxodnyx dannyx. At this
predobrabotki vÿpolnyayutsya avtomaticheski. For predobrabotki
program peredaetsya tolko otvet INS. Takim obrazom, prikladnoy
prikladnoy programmÿ.
modeling. Poetomu, o bolee spetsifichnoy predobrabotke doljna
The module
is installed on any computer, zadeystvovannom v
“Scheduler of tasks”
Scheme funktsionirovaniya module soglasuetsya so schemoy
The module organizes storage and access to neural network models. The model
consists of a set of INS and description sposoba predobrabotki vxodnyx
INS parameters.
"Planner task". Zadanie vklyuchaet v sebya odnu (esli operation
modeling. With kajdym proektom svyazano ne bolshe odnoy modeli. Proekt mojet ne
imet modeli, esli protsess modelirovaniya eshche ne
The module realizes the functions of pre-processing of prime. Predobrabotka
forms the
task of evolutionary structure, training and testing. Kriteriem selektsii
porojdaemyx neyrosetevyx modeley yavlyaetsya oshibka
gruppirovku (in frequency, clustering) primerov. Vse shagi
primeram osushchestvlyaetsya libo po kodu proekta, libo po kodu primera,
naznachaemogo prikladnoy programmoy. The first code has a strokovoy format
prikladnoy programmÿ model mojet bit vybrana iz bazy i vychislen
maximum number of synapses in INS. Zadaniya raspredelyayutsya po
postupleniya primera. Sposob formatirovaniya
stroki koda zavisit ot
vychislenie INS provoditsya na storone neyroservera. Prikladnoy
primenyayutsya metodÿ garantirovanno ne uxudshayushchie rezultatÿ
"Executor of the task"
pozabotitsya prikladnaya program.
"Base model"
program ne trebuetsya realizovÿvat protsedurÿ analiza struktury i
modelirovanii. “Performer of the task” performs the task, which is performed by
znacheniy. Takje imeyutsya svedeniya o nadejnosti models, poluchennÿe in xode
"Predobrabotchik"
evolyutsionnogo algorithm described in glave 2.3.3. Takje module
vÿpolnyalsya or he vÿpolnyaetsya in pervÿy raz i ne zavershen. On trebovaniyu
pozvolyaet dobavlyat, udalyat and zaprashivat primer. Obrashchenie k
vklyuchaet in sebya
zapolnenie spaces in dannyx, normalizatsiyu i
testing. On the structure of INS nakladÿvayutsya ogranicheniya nachalnogo i
113
Machine Translated by Google
Neurogenesis pozvolyaet vÿpolnyat parallelnuyu processing dannyx. V
contrasting structures. Dlya uskoreniya neyrovychisleniy
kompyuterax i podderjka mnogopotochnosti, realizuyut regime
v posledstvii izmenit. During the installation will be registered two services
gradients and stoxastic methods. Method obucheniya vybiraetsya
Do'stlaringiz bilan baham: 
