Federal gosudarstvennoe uchebnoe predpriyatie Chair of the System of Artificial Intelligence

Graphs of unipolar and bipolar sigmoidal functions
Koeffitsient b opredelyaet "krutiznu" funktsiy i vybiraetsya
or
ispolzuyutsya kak unipolyarnÿe, tak i bipolyarnÿe funktsii aktivatsii. Unipolar
function, as a rule, predstavlyaetsya formuloy
Graphic bipolar sigmoidal function
30
Machine Translated by Google

---

positive. Podobnaya nelineynaya characteristic reshaet postavlennuyu
izmenyaetsya ot malyx znacheniy pri bolshix otritsatelnyx vozbujdeniyax
tangent prinimaet znacheniya razlichnyx znakov, chto okazÿvaetsya vygodnÿm
vozbujdenii i snova umenshaetsya, when vozbujdenie stanovitsya bolshim
(see Fig. 3). In otlichie ot logisticheskoy funktsii giperbolicheskiy
By analogy with electronic systems activation function
obrabatyvat kak slabÿe, tak i silnye signalaly. Drugoy
shiroko ispolzuemoy aktivatsionnoy funktsiey yavlyaetsya
im dilemmu shumovogo nasÿshcheniya, thanks chemu odna i ta je set mojet
Proizvodnaya unipolyarnoy funktsii aktivatsii imeet vid -
for ryada zadach and prilojeniy.
mathematical model of activation of nerve cells: F = th
(x)
velichiny OUT k vÿzvavshemu ego nebolshomu prirashcheniyu velichiny NET. On
vÿrajaetsya naklonom krivoy pri opredelennom urovne vozbujdeniya i
(5)
schitaetsya nelineynoy usilitelnoy characteristic iskusstvennogo
giperbolicheskogo tangensa, chasto ispolzuetsya biologami v kachestve
neuron. Coefficient of usileniya vÿchislyaetsya as otnoshenie prirashcheniya
(krivaya pochti horizontalna) to maksimalnogo znacheniya pri nulevom
coordinate, i v tochke NET = 0 znachenie vyxodnogo signalala OUT ravno nulyu
Giperbolicheskiy tangens symmetrichen otnositelno nachala
31
Machine Translated by Google

---

imeyut ustoyavsheesya naznie mnogosloynyy perceptron MLP (MultiLayer Perceptron).
ot druga, i svoystva vsey seti ogranichenÿ svoystvami otdelnyx neyronov. Mnogosloynye
seti poluchili svoe razvitie s kontsa 70-x godov, kogda bÿli
15) tester
7) editor obuchayushchey vyborki
18) Contrast constructor
Odnonapravlennÿe mnogosloynye seti sigmoidalnogo tipi seti
6) zadachnik
Odnonapravlennÿe mnogosloynye neyronnye set sigmoidalnogo
Iz rassmotreniya konstruktsii neyroseti i osnovnyx rejimov ee
10) constructor otsenki