Federal gosudarstvennoe uchebnoe predpriyatie Chair of the System of Artificial Intelligence

2.1 Description of neural network technology
clustering - esli bez uchitelya (seti estestvennoy klassifikatsii);
drug from
another: 1) podgotovitelnye;
setey:
1) razdelenie na klassy ili klassifikatsiya, taksonomiya ili
2. Texnologii neyrosetevoy obrabotki dannyx
razdelit na neskolko grupp operatsiy, do nekotoroy stepeni nezavisimyx
nazÿvaemaya neyrosetevoy regressiey);
3) image recognition; 4) optimization;
5) forecasting; 6) modeling.
Additional tasks: 7) forecast («what
will be tomorrow?»); 8) conditional
forecast («chto budet zavtra,
esli ..?»); 9) determination of the
significance of the input parameters. Dlya resheniya
ukazannyx zadach razrabotan i ispolzuetsya
otsutstvuet algorithm or neizvestny printsipy resheniya zadach, no
matematicheskiy apparat iskusstvennyx neyronnyx setey, v chastnosti, iz
Neyrosetevye teknologii primenimy vo mnogix predmetnyx oblastyax, dlya
resheniya razlichnyx prikladnyx zadach. Est neskolko priznakov, kotorymi mojet ili
doljna obladat zadacha, chtoby primenenie neyronnyx
2) predskazanie (deystvitelnogo) chisla ili prediktsiya (inache
setey was justified:
reshat zadachi na osnove nakoplennogo opyta.
Neuroimitators - special programs for the creation, training, testing and other
operations with neural networks. Neuroimitators
test neyronnye seti, reshat razlichnÿe zadachi obrabotki dannyx.
sigmoidnyx neurons. Architecture and algorithms pozvolyayut obuchat i
51
Machine Translated by Google

---

3
2
field
operatsii (metodika podgotovki dannyx), nazvanie i naznachenie operatsiy, i
1
Ns
Thrombosis
Erythrocyte
3
primenyy diapazon znacheniy po kajdomu parameter, utochnit edinitsy
X1d Den
Numeric
X6 Leykotsity
signs
Ki
Numeric
Decryption
Numeric
dannyx (file dannyx) (primer - see table 2.1).
X2 Hemoglobin
4
parameters (ix korrelirovannost nesushchestvenna, v otlichie ot metodov
vanie
2) training of neural networks and other work with neural networks; 3) search results in
poluchennyx results. Nije privedeno kratkoe soderjanie grupp podgotovitelnyx
9
1
1
4
processing of various programs-neuroimitatorami. Determine
Numeric
3
p / p
3
Numeric
X3
In sootvetstvii s postanovkoy zadachi sleduet make a list
2
na
10
2
Numeric
dannye v vide ploskoy tablitsy, razrabotat strukturu tablitsy bazy
3
X1g God
3 X1m Months
X7 Eosinophils
X4 Reticulotsity
Naimeno
7
Numeric
Numeric
3
zapyatoy
ispolzovat latinskiy alfavit, chtoby obespechit sovmestimost pri
Donor code
kommentarii k nim. 1.
Sborn dannyx
Type Dli
X5
1
ÿ
izmereniya. Sobrat «syrye» dannye, esli oni eshche ne sobrany. Predstavit
4
Numeric
5
Numeric
Kol-vo
8
4
posle
factor analysis), dat short parameters of the name. Dlya imen luchshe vsego
11
Manipulation code
6
Table 2.1
Fragment structure of the basic data for the task of immune plasmapheresis
Numeric
52
Machine Translated by Google

---

On the practice of some characteristics of the object, presented in
12
Dbase and Paradox). Dlya relyatsionnyx baz dannyx ne trebuetsya daje provodit
Develop information classifiers (systems)
(antibody titer value up to 6 x103ME / l), medium (titer value from 6 x103ME / l to 12
x103ME / l) and high (titer value up to 24 x103ME / l). Diapazon razbivaetsya na tri
poddiapazona (kak ukazano). In sootvetstvii s
X8 Palochkoyadernye
obratnaya situation: inogda nujno provodit chastichnuyu or polnuyu
isxodnoy tablitsy, soderjashchee obÿchno nomer klassa. Classes predstavlyayut
sootvetstvie klassy (tselÿe chisla), opredelyaemÿe obÿchno v vide diapazonov
tselymi chislami - nomerami poddiapazona. Sootvetstvenno nekotorÿe
sgruppirovat by opredelennym criteria (svoystvam). Informatsionnye
2
klyuchevÿm polyam faylov dannyx v edinyy file (osobenno v sluchae
zadachax traditsionno dlya prakticheski lyubogo izuchaemogo parameter
Architecture sozdavaemoy base dannyx (relational, object-oriented i t. P.) In
obshchem sluchae ne imeet znacheniya, t. k. neyroimitator rabotaet obÿchno tolko s
odnim vxodnÿm faylom dannyx
dannyx i nujna v osnovnom dlya dalneyshey obrabotki v neyroimitatore, tak kak
neyroimitatory obÿchno rabotayut tolko s odnim vxodnym faylom
klassifikatsii (v otlichie ot prediktsii). Vozmojno ix ispolzovanie i dlya
ego change, kotoryye mojno predstavit in vide classes. For tasks
zavisimosti from konkretnoy zadachi.
Numeric
neuroimitatorov, razrabotannyx in Krasnoyarsk podderjivayut format
dannyx.
mojno sgruppirovat v tri gruppy po stepeni immunnogo otveta: nizkaya
13
etym razbieniem deystvitelnÿe znacheniya titer antibodies zamenyayutsya
normalization (v terminax teorii baz dannyx). The practice is typical
coding or rubrics). Dlya zadach klassifikatsii ispolzuetsya pole
vide konkretnyx znacheniy (deystvitelnyx chisel), mojno postavit v
soboy nekotorÿe sovokupnosti odnorodnyx objects, kotorye mojno
X9 Segmentoyadernye
denormalization, t. e. svedenie informatsii iz raznyx vzaimosvyazannyx po
izmeneniya znacheniy dannogo svoystva ob'ekta. For example, in medical
sushchestvuet diapazon ego izmeneniya, sootvetstvuyushchiy norme, libo urovni
3
bolshogo chisla tablits). Dannaya operation otnositsya k predobrabotke
klassifikatory poley dannyx ispolzuyut traditsionno dlya zadach
(imeet znachenie tolko format ispolzuemogo file dannyx: bolshinstvo
Numeric
dannyx. The structure of the end table is determined by the user in
zadach prediktsii, chto trebuet dopolnitelnyx operatsiy predobrabotki
immunosuppressive plasmapheresis analysis of donors based on antibody titer values
53
Machine Translated by Google

---

takoy podkhod primenen dlya opredeleniya professionalalnoy prinadlejnosti
mogut obuchitsya or pokazÿvayut neudovletvoritelnÿe rezultaty
posledovatelno, naprimer, ot menshix znacheniy k bolshim ili naoborot. Privedennyy
vyshe primer klassifikatsii donor po stepeni immunnogo
dayut nikakix preimushchestv (poryadok soxranyaetsya). Krome togo, v
high degree of immunization. V etom sluchae chislovye naimenovaniya
dopolnitelnogo preobrazovaniya polya, ispolzuyutsya znacheniya polya kak
kodirovaniya nikak ne meshaet, no i ne pomogaet pri reshenii zadachi. Danny
season year. At this, naturally, in bolshinstve sluchaev sushchestvenno
obÿekta, sleduet kajdoe sostoyanie kodirovat v otdelnom pole (naprimer, 1 -
nalichie dannogo sostoyaniya, 0 - otsutstvie). For example, it is characteristic
svoystv, esli eto vozmojno i esli izvestno kak. Takje vajno kodirovat
object (number, opredelyayushchie class, doljnÿ uvelichivatsya or
zapisey doljno bÿt bolshe kolichestva poley (parameters). Obychno pri
fill it as much as possible and tochnee.
immunizatsii, 2 - srednyaya, 3 - vysokaya. Dopustimo ispolzovat obratnuyu
volos, krome tsvetov radugi, kotoryye uporyadocheny po dline volny), inogda
setey oba privedennyx varianta kodirovaniya priznakov ekvivalentny i ne
nalichii takix poley). Inogda eto mojet pomoch, esli neyronnye seti ne
sostoyaniy (znacheniy) otdelnyx parameters of the object, kotoryye izmenyayutsya
gruppy). In dannyx primerax ne zadany stepen i otnosheniya poryadka, t. e. net
monotonnosti izmeneniya svoystv. For the task of immune plasmapheresis
testing of the use of neural networks "in the lobby" (without
otveta polnostyu sootvetstvuet dannomu opredeleniyu: nizkaya, srednyaya i
bolshinstve sluchaev mojno ispolzovat sleduyushchuyu klassifikatsiyu: 0 - nizkaya,
1- srednyaya, 2 - vysokaya (or obratnÿy poryadok znacheniy: 2 - nizkaya, 1 -
srednyaya, 0 - vÿsokaya). Obÿchno ispolzovanie nulevogo znacheniya dlya
donor (kajdaya profession kodiruetsya otdelnÿm polem), dney nedeli i
variant primenyayut for udobstva predobrabotki. Esli
je ne izvestna stepen uporyadochennosti znacheniy svoystva
est).
When compiling classifiers gelatelno monotonnoe change
klassov doljnÿ tochno otrajat poryadok sledovaniya sostoyaniy parameter
uvelichivaetsya chislo poley.
Gelatelno uchest sleduyushchee neglasnoe soglashenie: kolichestvo
odnim polem tablitsy bazy dannyx odin parameter. Sozdat bazu dannyx,
umenshatsya in sootvetstvii with sostoyaniyami). For example, 1 - low level
dlya parametrov, znacheniya kotoryx soderjat tsveta (naprimer, tsvet glaz ili
In chastnosti, monotonnoe change svoystv predpolagaet nalichie
zadachi prediktsii mojno predstavit kak zadachi klassifikatsii (pri
posledovatelnost: 1 - high, 2 - medium, 3 - low. For neurons
days of the week, seasons of the year, libo kakie-libo professii (professional
54
Machine Translated by Google

---

Esli simvolnÿe parameters imeyut kachestvennÿy smysl, takie
rassmotrena dalee (see p. 2.1.2). V etom sluchae kolichestvo zapisey ostaetsya
zapominaet svoe obuchennoe sostoyanie. Sushchestvuet neskolko sposobov
dannyy metod v neyroimitatore MultiNeuron udobnee.
zapisi. Posledniy sposob ne rekomenduetsya ispolzovat, tak kak on
uchityvat pri opredelenii otveta (for neyronnyx setey s uchitelem).
Pri nalichii v isxodnyx dannyx simvolnyx poley trebuetsya libo
izlojennyy in rabote podkhod. Kachestvennÿy parameter mojet otnositsya k
posledovatelnoe kodirovanie simvolnyx dannyx tselÿmy chislami. Pri
izvestnoy funktsiey raspredeleniya. Sleduet zametit, chto v nekotoryx
sluchaev. In neuroimitators MultiNeuron, naprimer, udobno ispolzovat
problemu nesbalansirovannosti kolichestva primerov v raznyx klassax, nejeli obshchego kolichestva
primerov. Esli net vozmojnosti sbora
primer (zapis) posledovatelno vybiraetsya dlya testirovaniya, i t. d., poka
podkhod, naprimer, razbit parameter na 2 ili bolee. Nezavisimo ot
nekotorÿe dopolnitelnÿe operatsii po razdeleniyu vyborok. Use
nablyudaetsya effect pereobucheniya: neyronnaya set v etom sluchae prosto
kontrastirovaniya (sokrashchenie chisla parametrov-poley), kotoraya budet
vozmojno.
izbejat etogo. For example, mojno provesti dopolnitelnÿy sbor dannyx. Some authors propose to add
to the database dannyx uje sushchestvuyushchie
tem je samÿm, no udalyayutsya naimenee znachimye polya, kotorye mojno ne
2. Ispolzovanie simvolnyx poley
parameters budem nazÿvat kachestvennymi, to mojno ispolzovat
otklyuchit ix, libo preobrazovat in chislovÿe. Obÿchno ispolzuetsya
fakticheski ne reshaet problemu nedostatochnosti dannyx. Tem ne menee, sushchestvuyut several
methods, osnovannyx na dannom podkhode. For example, for testing the work of algorithms
obucheniya ispolzuyut vÿborku, poluchennuyu iz ishodnoy s with the addition of some "noise" - signal
s
V raznyx neyroimitatorax sushchestvuyut razlichnÿe metodÿ dlya takix
odnomu iz chetyrex vidov:
neyroimitatorax pri nedostatochnom kolichestve dannyx (primerov vÿborki) uvelichivaetsya ves
otdelnyx klassov primerov. This is a characteristic, for example, for the neuroimitator MultiNeuron. The
application of this method reshaet skoree
«Skolzyashchiy kontrol»: training is conducted on the frequency of selection, odin
etom imeet znachenie poryadok numeratsii. Inogda trebuetsya bolee slojnÿy
dopolnitelnyx dannyx, inogda na praktike primenyayut protseduru
kolichestve zapisey sushchestvenno menshem, chem kolichestvo poley
neuroimitator ne pereberet vse vozmojnÿe zapisi. Strogo govorya, «skolzyashchiy kontrol» mojno
ispolzovat i v neyroimitatorax NeuroPro v. 0.25 in MDN, but when it comes to predvaritelno ispolzovat
konkretnyx variantov, prevrashchenie tekstovyx dannyx v chisla pochti vsegda
55
Machine Translated by Google

---

1) parameters, obladayushchie chastichnym
poryadkom; 2) parameters, obladayushchie otnosheniem sosedstva, no ne
obladayushchie poryadkom (dlya kajdogo elementa mojno ukazat sosedniy
element, no nelzya vÿdelit sredi nix predshestvuyushchiy);
4) parameters, yavlyayushchiesya kombinatsiey trex pervyx
vidov. Kak skazano ranee, rasprostranena praktika prisvoeniya vsem
sostoyaniyam poryadkovyx nomerov i podacha takogo psevdochislovogo parametra
na vxod neyronnoy seti. Pri etom seti navyazÿvaetsya isxodno otsutstvuyushchaya
information: razlichnÿm sostoyaniyam parameter pripisÿvayutsya otnosheniya
sosedstva i poryadka, a takje rasstoyanie mejdu vsemi sostoyaniyami.
Kachestvennÿy parameter kodiruetsya stolkimi vxodnymi signalalami
neyronnoy seti, skolko sostoyaniy on mojet prinimat. For kajdogo sostoyaniya
parameter opredelim ponyatiya verxnego konusa (mnojestvo vsex sostoyaniy
predshestvuyushchix dannomu sostoyaniyu) i sfer sosedstva radiusa 1, 2 i t. d. V
sferu sosedstva radiusa odin vxodyat vse sostoyaniya, yavlyayushchiesya sosedom
dannogo i ne svyazannÿe s nim otnosheniem predshestvovaniya. V sferu sosedstva
radiusa R vxodyat vse sostoyaniya, yavlyayushchiesya sosedyami sostoyaniy iz
sfery sosedstva radius R-1, ne svyazannÿe s dannym sostoyaniem otnosheniem
predshestvovaniya i ne prinadlejashchie sferam sosedstva sensstva radiusa. 3.
Vybor polya otveta After zaversheniya vsex operatsiy kodirovaniya dannyx sleduet
vybrat pole otveta libo polya otveta dlya mnogozadachnoy tablitsy dannyx ili
vektornogo prediktora. Esli obuchaetsya neyronnaya set s uchitelem, to
nujno vydelit pole otveta dlya neposredstvennogo ispolzovaniya blokom otsenki i
uchitelem pri obuchenii. Blok otsenki, sravnivaya izvestnoe zaranee znachenie iz
polya otveta dlya tekushchego primera («istinnyy» otvet) s vÿchislennÿm
znacheniem otveta, opredelyaet otsenku primera. On otsenke primerov
opredelyaetsya otsenka vÿborki i znachenie obratnogo signal. Esli obuchaetsya
neyronnaya set bez uchitelya (seti estestvennoy klassifikatsii), to nujno vydelit pole
(polya) otveta dlya vspomogatelnyx operatsiy libo dlya otdeleniya etogo polya ot
vxodnyx parametrov, t. k. pole otveta nelzya ispolzovat dlya razdeleniya
dannyx v kachestve vxodnogo
Ochevidno, chto sostoyaniya odnogo i togo je parametra mogut byt
uporyadocheny razlichnymi sposobami. Tak kak signal kodiruetsya dlya resheniya
konkretnoy zadachi, to neobxodimo rassmatrivat tolko te otnosheniya poryadka i
sosedstva, kotorye imeyut smÿsl v kontekste dannoy zadachi. Chtoby izbejat poteri
informatsii o parametre i v to je vremya ne nadelyat parameter ne prisushchimi
emu svoystvami, mojno ispolzovat sleduyushchiy sposob podachi kachestvennyx
priznakov.
3) parameters, ne obladayushchie ni odnim iz opisannyx vÿshe otnosheniy
(vse sostoyaniya yavlyayutsya izolirovannymi);
56
Machine Translated by Google

---

in some cases after processing neyronnymi setyami bez uchitelya
classification). For neyrosetey estestvennoy klassifikatsii pole otveta ne
srednim po klassu, esli klass izvesten;
srednim po polyu; nulem; minimum,
maximum. Vÿbor luchshego iz etix
sposobov - v zavisimosti ot tseny oshibki
ili
dalneyshem mojet slujit lish dlya proverki poluchennoy klassifikatsii, izvestnogo
raspredeleniya klassov - see. p. 2.1.4. Dlya dalneyshego ispolzovaniya neyronnyx
setey s uchitelem sleduet
blizkiy drugoy klass (esli kolichestvo klassov bolshe 2). Kolichestvo
FAMaster programs, Predmake and others.
Dannaya operation obÿchno gelatelna, but can not otsutstvovat pri
operatsiyu zapolneniya probelov kak operatsiyu modifikatsii dannyx daje
isklyuchaetsya iz list of vxodnyx parameters i ne podaetsya na vxod neyronnoy
poryadok. Gelatelno, chtoby dlya lyubyx i, j, gde i ÿ j (i, j - nomera klassov), iz
predicator. Krome togo, inogda mojno (nujno) vÿpolnit eksperimenty s
MultiNeuron) can be called primeram (klassam) ves. 4.
Poisk i zapolnenie probelov v dannyx Pri nalichii
probelov v dannyx sleduet zapolnit ix. Prosteyshie
dannyx budet tojdestvenno raven vxodnomu.
Zametim, chto znacheniya poley ravnye nulyu v obshchem sluchae ne yavlyayutsya
usage:
otsutstvovat, esli ispolzuyutsya neyronnÿe set bez uchitelya (estestvennoy
primerov lyubogo klassa doljno byt ne menshe 1, hotya kolichestvo 1 malo i
vÿzÿvaet problem, no inogda pri operatsx chteniya ili soxraneniya pustotÿ
ispolzuetsya pri obuchenii (esli takovoe bÿlo zaranee izvestno) i v
(estestvennoy klassifikatsii) dannyy klass mojet byt vklyuchen v naibolee
iz drugix soobrajeniy. For filling gaps can be used
uslovii zaranee izvestnoy polnotÿ dannyx. Mojno interpretirovat
from the list of fields select field answer. In any sluchae field answer
primerov raznyx klassov mojet byt neodinakovÿm, no imet odinakovÿy
seti. Dopustimo vybirat neskolko poley otveta dlya obucheniya vektornogo
posylki, chto Ni> Nj, sleduet Ni mojet byt okolo 2. V nekotoryx neyroimitatorax (naprimer
at isxodnoy zapolnennosti. In dannom sluchae modified kit
raznymi polyami answer.
Obuchayushchaya vÿborka doljna sostoyat iz primerov vsex klassov,
sostavlyayushchix klassifikatsionnuyu model. Drugimi slovami, kolichestvo
parameter. At the initial stage of field research the answer may be
sposoby zapolneniya probelov v dannyx po stepeni predpochtitelnosti ix
problems. Nulevÿe znacheniya ne yavlyayutsya pustymi. Obychno eto ne
57
Machine Translated by Google

---

set davat pravilnÿy otvet po obuchayushchey vÿborke. Classification
pered nachalom raboty. Frequently used normalization capabilities:
nekotorimi konstantami (naprimer, MultiNeuron zapolnyaet vse probely
experiments.
sleduet uchityvat pri rabote s dannymi, soderjashchimi probely.
srednee kvadratichnoe otklonenie, k - number sigm (usually 1, 2 or 3 in
strukturoy, where ukazÿvayutsya vxodnÿe, vyxodnÿe parameters, znachenie
etom dlya obucheniya ispolzuetsya dopolnitelnaya informatsiya - nomer klassa.
klassifikatsii), number of layers, number of neurons in each layer.
kvadratichnoe otklonenie, parnÿe korrelyatsii, korrelyatsionnoe otnoshenie.
normirovaniya budet diapazon [a, b], gde a <0, b> 1, pryamaya otobrajaetsya v
normally. Krome togo, dannyx chasto nedostatochno dlya obosnovaniya
na interval [0,1] nelineyno, s ispolzovaniem, naprimer, sigmoidnogo
teacher
on the sphere of the same radius (kajdyy vector delitsya on its length). Sposob
normirovaniya mojet byt izmenen posle provedeniya ryada
dannyx. Some neuroimitators automatically fill the void
6. Normirovanie dannyx. Obÿchno vÿpolnyaetsya neuroimitatorom
neyronnyx setey s uchitelem. Teacher - block, which obuchaet neyronnuyu
znacheniem «-1») ili raschetnymi znacheniyami po nekotorym pravilam. This
at interval [0,1] linearly; at
interval [0,1] kusochno lineyno [-ks, + ks], where s est vyborochnoe
In konechnom itoge sozdaetsya neskolko neyrosetey so sxodnoy
neyronnymi setyami s uchitelem obÿchno vÿpolnyaetsya ÿffektney, t. k. pri
otkloneniya (accuracy) znacheniya vyxodnoy velichiny (for zadach prediktsii) libo znacheniya
vyxodnoy velichiny for kajdogo klassa (zadachi
5. Definition of simple statistical parameters: average, average
sootvetstvii s prinyatymi v statistike), na levuyu granitsu otobrajaetsya vse, chto levee, na
pravuyu granitsu - vse, chto pravee; at interval [0,1] nestrogo, not linear [-ks, + ks], t. e. result
Proofka gipotezy o normalnom raspredelenii dannyx vyborki, t. k. dlya dannyx, sobrannyx
v realnyx usloviyax, raspredelenie daleko ne vsegda
pryamuyu, no interval [-ks, + ks] otobrajaetsya in [0, 1];
2.1.2. Description of technologies used in neural networks s
normalnosti raspredeleniya.
v chislovyx polyax avtomaticheski zapolnyayutsya nulevymi (ili inymi) znacheniyami v nekotoryh
neyroimitatorax. It sleduet uchityvat, t. k. dannaya nastroyka mojet sushchestvenno povliyat na
rezultaty obrabotki
normalization;
Dannÿe s izvestnÿm otvetom mojno obrabatÿvat s pomoshchyu
58
Machine Translated by Google

---

ispolzovavshixsya for training. Inymi slovami, testirovanie - check
setey s uchitelem obrabotat nevozmojno. Dlya obrabotki takix dannyx
solution options. Obuchennaya set on vxodnÿm dannÿm, analogichnym po
translucent neural network; 3)
receipt of verbal description of the set; 4)
testing obuchennoy seti; 5) first training.
Rassmatrivaya neyronnuyu set as a set of
elements, proizvodyashchix
opredelennyy answer (class).
In nastoyashchee vremya osnovnoy dlya neyroimitatorov na PK nazÿvayut
klassifikatsii (unsupervised by zarubejnoy klassifikatsii neyronnyx
reshenie glavnoy i vspomogatelnyx zadach. In the main task vÿdelyayut
setey. Generation - the task of nachalnyx znacheniy all neyronnoy parameters
otsutstviya klassifikatsionnoy model, chto chasto bÿvaet na nachalnom etape
peredano ot znayushchego subÿekta k poznayushchemu v xode obÿchnogo akta
klassifikatsii) prednaznacheny dlya postroeniya klassifikatsionnoy modeli, a takje
pozvolyayut opredelyat znachimosti vxodnyx signalov i
sotsialno-kulturnoy sredÿ, traditsiy, context obshcheniya.
Neyronnaya set delitsya s nami znaniyami, tolko esli my ee etomu
uchitelem tak, chtobÿ neyronnaya set po obuchayushchey vÿborke davala
teacher):
1) definition of significance of input parameters; 2)
contrast of the neural network or semiconductor logic
number of classes), to takie dannÿe bez izvestnogo otveta s pomoshchyu neyronnyx
yavlyaetsya ukazanie klassa - vybor odnogo iz neskolkix vozmojnyx
seti s izvestnymi otvetami po testovoy vyborke - naboru primerov, ne
ispolzuyut bolee prostoy instrument - neyronnÿe seti estestvennoy
strukture obuchayushchey vÿborke, doljna bÿt sposobna vÿchislyat
nekotorÿe vÿchisleniya nad prixodyashchimi k nim dannymi, mojno izuchat
kachestva obucheniya seti. Vspomogatelnÿe tasks: opredelenie znachimosti
funktsii ili podzadachi: generatsii, obucheniya i testirovaniya neyronnyx
setey), takje nazÿvaemÿe samoorganizuyushchiesya seti. They work in sluchae
zadachu «izvlechenie znaniy iz dannyx». Yavnoe znanie - to, chto mojet byt
lyubogo issledovaniya. Neyronnye seti uchitelya (estestvennoy)
obshcheniya (kommunikatsii) between subjects. Yavnoe knowledge depends on
seti; obuchenie - izmenenie vsex ili chasti parametrov neyronnoy seti s
minimizirovat ix kolichestvo. Iz
vsex tipov zadach, reshaemyx na osnove nakoplennogo ranee opyta,
znachitelnuyu gruppu sostavlyayut zadachi klassifikatsii. Answer for nix
Odnako, esli nomer klassa zaranee neizvesten (chasto bÿvaet neizvestno i
nauchili. Osnovnÿe sposoby polucheniya yavnogo znaniya ot neyronnoy seti (s
required answers; testirovanie - sravnenie otvetov obuchennoy neyronnoy
59
Machine Translated by Google

---

trebuemÿy answer. Vspomogatelnÿe tasks sformulirovany in the process
over Microsoft Excel.
Sleduet otmetit, that some neuroimitators podderjivayut
neyronnye seti, podderjivayut v osnovnom formaty xls i nekotorye
potrebnosti users and napravleniya razvitiya. Namely
standartizovannuyu and strogo unikalnuyu. This strongly zatrudnyaet rabotu s
save to my office packages (Microsoft Excel, Word) or internally
formats Dbase, Paradox (file dbf, db), and the result of the chashche vsego legko
resheniya prikladnyx zadach polzovatelya.
Obuchenie neyronnyx setey s uchitelem mojet byt vÿpolneno s
other programmnÿe produkty dlya obrabotki poluchennyx rezultatov
mnogix drugix. Krome togo, in nastoyashchee vremya in nekotoryx paketax
Odnako spravedlivo i to, chto sushchestvuyut neyroimitatory, podderjivayushchie
obshcheprinyatye formaty dannyx, libo pozvolyayushchie
soderjat naimenovaniya poley (gelatelno ispolzovanie latinskix bukv
obshcheprinyatyx. For example, statistical packages, pozvolyayushchie realizovat
primera, chtoby obuchennaya neyronnaya set pri testirovanii davala
modules for working with neural networks. For example, products of the series Excel
Neural Package (Winnet, Kohonen map), realizovannye kak nabor nadstroek
vyxodnÿe polya vo vxodnyx faylax dannyx doljny byt chislovogo tipa,
razvitiya i prakticheskogo ispolzovaniya neyronnyx setey, kak novye
sobstvennÿe vnutrennie formaty, kak vxodnyx, tak i vyxodnyx dannyx, libo je
predpolagayut ix opredelennuyu strukturu zachastuyu ne
data base formats (Dbase, Paradox), processing results can be transferred
format. Neuroimitators, developed in g. Krasnoyarsk (MultiNeuron, NeuroPro, MDN), in
kachestve vxodnyx dannyx obÿchno podderjivayut
vspomogatelnÿe zadachi chashche vsego est osnovnoy element teknologii
podobnymi neyroimitatorami i ne pozvolyaet ÿÿÿÿÿÿÿÿÿÿ ÿÿÿÿÿÿÿÿÿÿ
pomoshchyu programm Clab [S. E. Gilev], MultiNeuron [A. A. Rossiev, E. M. Mirkes, S.
E. Gilev et al.], NeuroPro [V. G. Tsaregorodtsev], Monoton i
(problem of compatibility of formats).
konvertiruyutsya in tekstovÿy format (txt). When using Microsoft Excel spreadsheet data
in xx format you can easily convert to Dbase format. At the same time, the dollar should
be formed as a table in the first stroke
statistics i in ÿÿÿÿÿÿÿÿÿÿÿ ÿÿÿÿÿÿÿÿ predusmotrenÿ dopolnitelnÿe
vxodnyx signals, kontrastirovanie neyronnyx setey, obuchenie primera, opredelenie
minimalnogo reshayushchego nabora vxodnyx parametrov, poluchenie logicheski
prozrachnoy neyronnoy seti i, v konechnom schete, znaniy iz dannyx. Prime training -
such a change of parameters
konvertirovat (for vyxodnyx dannyx) svoy vnutrenniy format v odin iz
vo izbejanie problem sovmestimosti mejdu versiyami). Vxodnye i
60
Machine Translated by Google

---

1. Obuchenie neyronnoy seti po polnoy vyborke. Neyronnaya set s
otvetom, a takje grupp primerov s bolee slojnoy strukturoy protivorechiya
s oshibochno opredelennoy klassifikatsiey pri testirovanii).
vozrastaet i mogut poyavitsya protivorechivye primers. Protivorechivost
soderjimogo kajdogo iz klassov (rassmatrivaetsya kajdyy primer i
protivorechie pri dopolnitelnom nabore dannyx, kogda obÿem vÿborki
neprotivorechivosti dannyx, otsutstvii odinakovyx primerov s raznÿm
sravnivaetsya s sootvetstvuyushchey klassifikatsiey: osobo izuchayutsya primerÿ
opisanii predobrabotki (see p. 2.1.1).
uchten odin ili neskolko faktorov (vxodnyx poley), sushchestvenno
(Fig. 2.1).
primernaya posledovatelnost preobrazovaniya poley privedena pri
mojet bÿt poluchena v isxodnoy tablitse dannyx, esli issledovatelem ne
priznak protivorechiya dannyx est priznak nepolnoty ili nedostatochnosti
Consider the structure of the protivorechiya in the first biological task
kommentarii k nim.
vliyayushchix na izuchaemÿy protsess v rassmatrivaemoy oblasti znaniy. Inogda
Nije privedeno kratkoe opisanie grupp osnovnyx operatsiy dlya
neyrosetevoy teknologii s uchitelem, ix nazvanie, naznachenie i
posle udaleniya odnogo ili neskolkix vxodnyx parametrov, daje esli
isxodno ona bÿla neprotivorechiva (tabl. 2.2). Mojno poluchit
uchitelem, obuchennaya po polnoy vyborke, svidetelstvuet o prosteyshey
nablyudaetsya, no mojet bÿt vÿyavleno posle detalnogo analiza
classification of iris Fishera. Protivorechivost dannyx mojet poyavitsya
klassifikatsionnoy model. Yavnogo protivorechiya v etom sluchae iskhodno ne
sostoyanie
sostoyanie
sostoyanie
sostoyanie
sostoyanie