Perseptron. ko’p qatlamli perseptron
![I.BOB. SUNIY INTELLEKT FANI
1.1. Suniy intellekt haqida
Sun iy ong, sun iy intellekt yoki sun iy idrok (inglizcha: Artificialʼ ʼ ʼ
intelligence; odatda, AI sifatida ham qisqartiriladi) — insonlar yoki hayvonlar
tomonidan ko rsatiladigan tabiiy ongdan farqli o laroq, mashinalar tomonidan
ʻ ʻ
ko rsatiladigan ongdir. Yetakchi sun iy ong darslik kitoblari bu sohani „ongli
ʻ ʼ
agentlar“ni o rganish deya ta riflaydi: o z muhitini fahmlaydigan va maqsadlariga
ʻ ʼ ʻ
muvaffaqiyatli erishish imkoniyatini maksimal darajada oshiradigan amallarni
amalga oshiruvchi har qanday sistema. Xalq orasida „sun iy ong“ atamasi
ʼ
ko pincha „o rganish“ va „muammolarni yechish“ kabi inson idroki bilan
ʻ ʻ
bog laydigan „kognitiv“ funksiyalarni taqlid qiladigan mashinalarni tasvirlashda
ʻ
ishlatiladi, biroq bu ta rifni yirik sun iy ong tadqiqotchilari rad etishadi
ʼ ʼ 1
.
Sun iy ong ilovalari yetuk web-qidiruv tizimlari (masalan, Google), tavsiya
ʼ
etuvchi tizimlar (bundan YouTube, Amazon va Netflix foydalanadi), inson nutqini
anglash (masalan, Siri yoki Alexa), o ziyurar mashinalar (masalan, Tesla) hamda
ʻ
strategik o yin tizimlarida (masalan, shaxmat va Go) yuqori darajada
ʻ
raqobatlashishni o z ichiga oladi.[1] Mashinalar tobora ko p qobiliyatlarga ega
ʻ ʻ
bo lib borishar ekan, „ong“ talab etuvchi vazifalar ko pincha sun iy ong effekti deb
ʻ ʻ ʼ
ataluvchi fenomen bo lgan sun iy ong ta rifidan olib tashlanadi.
ʻ ʼ ʼ
Sun'iy intellekt" atamasi 1956 yilda paydo bo'lgan, ammo bugungi kunda SI
texnologiyasi ma'lumotlar hajmini ko'paytirish, algoritmlarni takomillashtirish,
hisoblash quvvatini va ma'lumotlarni saqlash vositalarini optimallashtirish fonida
haqiqiy mashhurlikka erishdi.
O'tgan asrning 50-yillarida boshlangan SI sohasidagi birinchi tadqiqot
muammolarni hal qilish va ramziy hisoblash tizimlarini rivojlantirishga qaratilgan
edi. 60-yillarda bu sohada AQSh Mudofaa vazirligi qiziqish uyg'otdi: AQSh
harbiylari insonning aqliy faoliyatini simulyatsiya qilish uchun kompyuterlarni
o'qitishni boshladi.
1
Linnainmaa, Seppo (1970). The representation of the cumulative rounding error of an algorithm as a Taylor
expansion of the local rounding errors (Masters) (in Finnish). University of Helsinki. pp. 6–7.
5](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_4.png?v=1)
![Ushbu bo'limda biz Iris ma'lumotlar to'plamidagi gullarni tasniflash uchun
Python-da oddiy perceptron o'rganish qoidasini amalga oshiramiz.
A ko'p qatlamli pertseptron (MLP) a noto'g'ri nom zamonaviy uchun ozuqa
oldinga sun'iy neyron tarmoq, kamida uchta qatlamda tashkil etilgan, chiziqli
bo'lmagan faollashtirish funktsiyasiga ega bo'lgan to'liq bog'langan neyronlardan
iborat bo'lib, chiziqli ravishda ajratib bo'lmaydigan ma'lumotlarni ajrata olishi
bilan ajralib turadi.[1] bu noto'g'ri nom, chunki asl perceptron a Heaviside qadam
funktsiyasi, chiziqli bo'lmagan faollashtirish funktsiyasi o'rniga (zamonaviy
tarmoqlar tomonidan ishlatiladi).
Zamonaviy besleme tarmoqlari backpropagation usuli yordamida o'qitiladi
va og'zaki ravishda "vanil" neyron tarmoqlari deb ataladi.
Timeline
1958 yilda kirish qatlamidan, o'rganmagan randomizatsiyalangan
og'irliklarga ega yashirin qatlamdan va o'rganish ulanishlari bilan chiqish
qatlamidan iborat bo'lgan pertseptronlarning qatlamli tarmog'i allaqachon kiritilgan
Frank Rozenblatt uning kitobida Pertseptron.[8][9][10] bu ekstremal ta'lim
mashinasi[11][10] hali chuqur ta'lim tarmoq emas edi.
21](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_20.png?v=1)
![1965 yilda Stokastik gradient tushishidan hali foydalanmagan birinchi
chuqur o'rganish tarmog'i Aleksey Grigorevich Ivaxnenko va Valentin Lapa
tomonidan nashr etilgan.ma'lumotlar bilan ishlashning guruh usuli 4
.
1967 yilda birinchi marta stoxastik gradient tushishidan foydalangan,
chiziqli bo'lmagan ajratiladigan naqsh sinflarini tasniflashga qodir bo'lgan chuqur
o'rganish tarmog'i Shun'ichi Amari tomonidan nashr etildi.[14] Amarining talabasi
Saito ikkita o'quv qatlami bo'lgan besh qatlamli besleme tarmog'idan foydalangan
holda kompyuter tajribalarini o'tkazdi.
1970 yilda zamonaviy backpropagation usuli, zanjir qoidalariga asoslangan
nazorat ostida o'rganishni samarali qo'llash,[15][16] birinchi marta fin
tadqiqotchisi Seppo Linnainmaa tomonidan nashr etilgan.[2] [17] [10] atama (ya'ni
"orqaga tarqaladigan xatolar") o'zi Rozenblattning o'zi tomonidan ishlatilgan, lekin
u buni qanday amalga oshirishni bilmas edi, garchi doimiy kashshof bo'lsa ham
backpropagatsiya kontekstida allaqachon ishlatilgan boshqaruv nazariyasi yilda
1960 Genri J. Kelley tomonidan. bu avtomatik farqlash teskari rejimida sifatida
ham tanilgan.
1982 yilda backpropagation birinchi marta Pol Verbos tomonidan standartga
aylangan tarzda qo'llanildi.
1985 yilda texnikaning eksperimental tahlili Devid E. Rumelxart va boshq..
[6] yondashuv uchun ko'p yaxshilashlar keyingi o'n yilliklar ichida qilingan, [10].
1990-yillarda neyron tarmoqlardan foydalanishga (ancha sodda) alternativa,
garchi hali ham bog'liq bo'lsa ham[18] quvvatlash vektorli mashina yondashuvi
Vladimir Vapnik va uning hamkasblari tomonidan ishlab chiqilgan. Chiziqli
tasniflashni amalga oshirishdan tashqari, ular yuqori o'lchovli xususiyat
bo'shliqlaridan foydalangan holda yadro hiylasi deb ataladigan chiziqli bo'lmagan
tasnifni samarali bajarishga muvaffaq bo'lishdi.
4
Rosenblatt, Frank (1958). "The Perceptron: A Probabilistic Model For Information Storage And Organization in
the Brain". Psychological Review. 65 (6): 386–408. CiteSeerX 10.1.1.588.3775. doi:10.1037/h0042519. PMID
13602029. S2CID 12781225.
22](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_21.png?v=1)
![2003 yilda Yoshua Bengio tomonidan hammualliflar bilan tilni
modellashtirishda chuqur o'rganish yutuqlari tufayli backpropagatsiya tarmoqlariga
qiziqish qaytdi.
2017 yilda zamonaviy transformator arxitekturalari joriy etildi.
2021 yilda ikkita chuqur Mlpni skip ulanishlari va qatlam normalizatsiyasi
bilan birlashtirgan juda oddiy NN arxitekturasi ishlab chiqilgan va MLP-mikser
deb nomlangan; 19 dan 431 milliongacha parametrlarga ega bo'lgan uning amalga
oshirilishi ImageNet-da o'xshash o'lchamdagi ko'rish transformatorlari va shunga
o'xshash tasvirlarni tasniflash vazifalari bilan taqqoslanishi ko'rsatilgan.
Quyidagi misol uchun biz Iris ma ' lumotlarini uci Machine Learning
omboridan yuklaymiz va faqat Setosa va Versicolor ikkita gul turiga e ' tibor
qaratamiz . Bundan tashqari , biz vizualizatsiya maqsadida sepal uzunligi va gulbarg
uzunligining ikkita xususiyatidan foydalanamiz .
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
from mlxtend.plotting import plot_decision_regions
df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/
iris/iris.data', header=None)
### select training data and identify label
y=df.iloc[0:100,4].values
y=np.where(y=='Iris-setosa',-1,1)
x=df.iloc[0:100,[0,2]].values
###now we define the necessary functions, especiall the fit()
class Perceptron():
"""Perceptron classifier.
Parameters
------------
eta : float
Learning rate (between 0.0 and 1.0)
23](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_22.png?v=1)
![n_iter : int
Passes over the training dataset.
Attributes
-----------
w_ : 1d-array
Weights after fitting.
errors_ : list
Number of misclassifications in every epoch.
"""
def __init__(self, eta=0.01, n_iter=10):
self.eta = eta
self.n_iter = n_iter
def fit(self, x, y):
"""Fit training data.
Parameters
----------
X : {array-like}, shape = [n_samples, n_features]
Training vectors, where n_samples
is the number of samples and
n_features is the number of features.
y : array-like, shape = [n_samples]
Target values.
Returns
-------
self : object
"""
self.w_ = np.zeros(1 + x.shape[1])
self.errors_ = []
for _ in range(self.n_iter):
errors = 0
24](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_23.png?v=1)
![for xi, target in zip(x, y):
update = self.eta * (target - self.predict(xi))
self.w_[1:] += update * xi
self.w_[0] += update
errors += int(update != 0.0)
self.errors_.append(errors)
return self
def net_input(self, x):
"""Calculate net input"""
return np.dot(x, self.w_[1:]) + self.w_[0]
def predict(self, x):
"""Return class label after unit step"""
return np.where(self.net_input(x) >= 0.0, 1, -1)
pctrn = Perceptron(n_iter=10, eta=0.1)
pctrn.fit(x,y)
plot_decision_regions(x, y, clf=pctrn)
plt.title('Perceptron')
plt.xlabel('sepal length [cm]')
plt.ylabel('petal length [cm]')
plt.show()
plt.plot(range(1,len(pctrn.errors_)+1),pctrn.errors_,marker='o')
plt.xlabel('Epochs')
plt.ylabel('Number of misclassifications')
plt.show()
25](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_24.png?v=1)
![FOYDALANILGAN ADABIYOTLAR
1. Cybenko, G. 1989. Approximation by superpositions of a sigmoidal
function Mathematics of Control, Signals, and Systems, 2(4), 303–314.
2. Linnainmaa, Seppo (1970). The representation of the cumulative
rounding error of an algorithm as a Taylor expansion of the local rounding errors
(Masters) (in Finnish). University of Helsinki. pp. 6–7.
3. Kelley, Henry J. (1960). "Gradient theory of optimal flight paths".
ARS Journal. 30 (10): 947–954. doi:10.2514/8.5282.
4. Rosenblatt, Frank. x. Principles of Neurodynamics: Perceptrons and
the Theory of Brain Mechanisms. Spartan Books, Washington DC, 1961
5. Werbos, Paul (1982). "Applications of advances in nonlinear
sensitivity analysis" (PDF). System modeling and optimization. Springer. pp. 762–
770. Archived (PDF) from the original on 14 April 2016. Retrieved 2 July 2017.
6. Rumelhart, David E., Geoffrey E. Hinton, and R. J. Williams.
"Learning Internal Representations by Error Propagation". David E. Rumelhart,
James L. McClelland, and the PDP research group. (editors), Parallel distributed
processing: Explorations in the microstructure of cognition, Volume 1:
Foundation. MIT Press, 1986.
7. Hastie, Trevor. Tibshirani, Robert. Friedman, Jerome. The Elements
of Statistical Learning: Data Mining, Inference, and Prediction. Springer, New
York, NY, 2009.
8. Rosenblatt, Frank (1958). "The Perceptron: A Probabilistic Model For
Information Storage And Organization in the Brain". Psychological Review. 65
(6): 386–408. CiteSeerX 10.1.1.588.3775. doi:10.1037/h0042519. PMID
13602029. S2CID 12781225.
9. Rosenblatt, Frank (1962). Principles of Neurodynamics. Spartan, New
York.
10. Schmidhuber, Juergen (2022). "Annotated History of Modern AI and
Deep Learning". arXiv:2212.11279 [cs.NE].
27](https://docx.uz/documents/37c0d5cc-c551-48ad-bc5d-dead5529190d/page_26.png?v=1)
PERSEPTRON. KO’P QATLAMLI PERSEPTRON