Sun’iy intellekt

• Neyron tarmoqlari

• Adabiyotlar

Sun’iy intellekt

• Yigirmanchi asrda sun'iy intellekt tushunchasi shakllantirildi va bu tushunchaning bir qator ta'riflari taklif qilindi. Ko'pchilik tomonidan qabul qilingan va bugungi kunda mashhur bo'lgan birinchi ta'riflardan biri bu: "Kompyuterlarni odamlar kabi fikrlashga majbur qilish usuli".

• Ushbu ta'rifning kelib chiqishi ingliz matematiki va dastlabki kompyuter fanlari tadqiqotchisi Alan Turing tomonidan taklif qilingan mashhur test edi. Ushbu testda eksperimentator masofaviy klaviatura yordamida xabar almashadigan "odam" shaxs yoki kompyuter dasturi ekanligini aniqlashga harakat qiladi.

• Suniy intellekt

• Sun'iy intellekt sohalari

• Sun’iy intellekt

• Sun’iy intellekt

• Suniy neyron
• tarmoqlari

• Tanib olishga eng oddiy vazifa

• 1998-yil
• MNIST ma'lumotlar bazasida 60 000 ta qo'lda yozilgan raqamli tasvirlar va 10 000 ta sinov tasvirlari mavjud.
• 2010 yil
• 0,27% xatolik bilan CNN

• MNIST - Mixed National Institute of
• Standards and Technology

• https://en.wikipedia.org/wiki/MNIST_database

• http://yann.lecun.com/exdb/mnist/

• Neyron tarmoqlari

• Rstudio da MNIST
• 1-usul

• Neyron tarmoqlari

• Rstudio da MNIST
• 2-usul

• install.packages("darch")
• library("darch", lib.loc="/Library/Frameworks/R.framework/Versions/3.1/Resources/library")
• readMNIST("/Users/felixfilippov/Desktop/mnist_fold/")
• [1] "Loading the the MNIST data set."
• [1] "Loading train set with 60000 images."
• [1] "Generating randomized data set and label matrix"
• [1] "class 0 = 5923 images"
• [1] "class 1 = 6742 images"
• [1] "class 2 = 5958 images"
• [1] "class 3 = 6131 images"
• [1] "class 4 = 5842 images"
• [1] "class 5 = 5421 images"
• [1] "class 6 = 5918 images"
• [1] "class 7 = 6265 images"
• [1] "class 8 = 5851 images"
• [1] "class 9 = 5949 images"
• [1] "Saving the test data (filename=train)"
• [1] "Loading test set with 10000 images."
• [1] "Generating randomized data set and label matrix"
• [1] "class 0 = 980 images"
• [1] "class 1 = 1135 images"
• [1] "class 2 = 1032 images"
• [1] "class 3 = 1010 images"
• [1] "class 4 = 982 images"
• [1] "class 5 = 892 images"
• [1] "class 6 = 958 images"
• [1] "class 7 = 1028 images"
• [1] "class 8 = 974 images"
• [1] "class 9 = 1009 images"
• [1] "Saving the test data (filename=test)"
• [1] "Finished"

• Neyron tarmoqlari

• Oxir-oqibatda:
• 99,67% aniqlik bilan tanib olish. 10 000 ta sinovdan faqat 33 tasi xato bilan tasniflanadi.
• Bu yerda xatolar ko’rsatilgan:
• yuqori o'ng burchakdagi qiymat MNIST bo'yicha to'g'ri tasnif, pastki o'ng burchakdagi qiymat esa noto'g'ri tasnifdir.

• Oddiy tanib olish vazifasi

• Neyron tarmoqlari

• Rosenblatt Perceptron.

• y =

• 0 agar Σxiwi ≤ chegara
• 1 Agar Σxiwi > chegara

• x1

• x2

• x3

• y

• w1

• w2

• w3

• y =

• 0 agar X W + b ≤ 0
• 1 Agar X W + b > 0

• x1

• x2

• x3

• y

• w1

• w2

• w3

• b

• .

• .

• x1

• x2

• -2

• 3

• -2

• 00 0x(-2) + 0x(-2) + 3 = 3 > 0 y = 1
• 01 0x(-2) + 1x(-2) + 3 = 1 > 0 y = 1
• 10 1x(-2) + 0x(-2) + 3 = 1 > 0 y = 1
• 11 1x(-2) + 1x(-2) + 3 = -1 < 0 y = 0

• y

• http://neuralnetworksanddeeplearning.com/movies/create_step_function.mp4

• Neyron tarmoqlari

• http://habrahabr.ru/post/214109/

• Inson miyasi ko'p sonli ulanishlar bilan bog'langan taxminan 86 milliard neyrondan iborat (har bir neyronga o'rtacha bir necha ming ulanish, lekin bu raqam juda o'zgarishi mumkin).

• Miya

• http://habrahabr.ru/post/108483/

• Neyron tarmoqlari

• Neyron

• Neyronlar:
• ko’p kirishlar (
  
  Download 22.25 Mb.