Federal gosudarstvennoe uchebnoe predpriyatie Chair of the System of Artificial Intelligence

Simple summator (Figure 1) does not have the specified parameters. On imeet
vektornyy vxod i vydaet na vyxode summu koordinat vxodnogo
adaptive summatora, imeyushchego n + 1 input and poluchayushchego on 0-y input
Ochen vajnyy element neyrosetey - eto adaptivnyy summator. Adaptive
summator calculates the scaling of the input vector
postoyannuyu. Ispolzovanie edinitsy chashche vsego udobnee, hotya ne
synaptic coefficient ÿ. The synapse receives a scaly input
odnorodnuyu funktsiyu (x, ÿ), has n nastraivaemyx parameters. Na schemax
neyroinformatike idealnaya schemotexnika predstavlyaet soboy osobyy yazyk
heodnorodnyy adaptivnyy summator (ris. 3). In
some tasks mogut rassmatrivat and other types of summators, for example,
kvadratichnyy. Kvadratichnÿy summator vÿchislyaet kvadratichnuyu
summatorov sinaps obÿchno ne vstrechaetsya, odnako dlya nekotoryx
20
Machine Translated by Google

---

poluchaet skalyarnÿy vxodnoy signal x i perevodit ego in ÿ (x). Function ÿ
linear function. Funktsiya aktivatsii mojet by stupenchatoy (sign) or
element, nazÿvaemÿy vyxodnaya zvezda. At the entrance he receives a signal
(activation function). Kak pravilo, ona nelineyna, no inogda mojet byt i
signalov, inogda takie bloki daje ne schitayut neyronami.
Sometimes it is useful to connect synapses and tochki vetvleniya, obrazuya
diapazon izmeneniya vesov synapsov nakladÿvayutsya nekotorye ogranicheniya,
naprimer, prinadlejnosti vesa sinapsa diapazonu [-1, 1] ili drugoy
Tochka vetvleniya slujit dlya rassylki odnogo signalala po neskolkim
turn on the signal distribution. Analogichno etomu tochka vetvleniya
gladkoy. Sredi gladkix functional chashche vsego ispolzuyutsya sigmoidnÿe.
Podrobnee nekotorÿe vidy funktsiy aktivatsii obsujdayutsya nije.
nastraivaya kotoryye, neyronnaya set obuchaetsya resheniyu task. Obychno na
(A), na k svoix vyxodax vydaet signalaly ÿiA (i = 1,…, k). Vyxodnaya zvezda -
element, dvoystvennÿy adaptivnomu summatoru, - perexodit v nego, esli
vsem svoim vyxodam.
Formal neuron chashche vsego sostoit iz vxodnogo summatora,
nelineynogo preobrazovatelya i tochki vetvleniya na vyxode (ris. 8).
Sushchestvuyut architecture neyronnyx setey, v kotoryx neyron imeet dva
Nelineynyy preobrazovatel signala izobrajen na ris. 6. On
summatora - neuron Pade. Vozmojny takje i drugie tipy neyronov i
address (Fig. 7). Ona poluchaet skalyarnÿy vxodnoy signal x i peredaet ego
comfortable range.
Adaptive summator can be represented by using n
linear ligaments and simple summators (Fig. 5).
double prostomu summatoru.
prinyato nazÿvat function aktivatsii or characteristic feature
opredelyaet maksimalnÿy (or minimalnyy) iz prishedshix k nemu
summatorov, naprimer, neyron vtorogo sloya v nekotoryx setyax Koxonena
Rice. 1. Simple summator
21
Machine Translated by Google

---

ÿ
x1
2
0
x2
1
n
xn
ÿ
,
ÿ
ÿ
x
ÿ
ÿ
ÿ
.
.
.
1
Rice. 4. Synapse or svyaz
Rice. 2. Adaptive summator
Rice. 3. Unusual adaptive summator
in prostogo summatora
Rice. 5. Adaptive summator, presented as a composition n lineynyx svyazey
22
1
2
n
ÿ
ÿ
ÿ
ÿ
ÿ
ÿ
0 ÿ ÿ ÿ x
ÿ
ÿ
,
...
ÿ
ÿ
signal
Vxodnoy
xi i
signal
ÿ
n
Vyxodnoy
i
1
ÿ ÿ x