Ɗaya daga cikin sanannun kayan aikin don haɓaka ƙirar koyon injin shine TensorFlow. Muna amfani da TensorFlow a aikace-aikace da yawa a masana'antu daban-daban.
A cikin wannan sakon, za mu bincika wasu samfuran TensorFlow AI. Don haka, zamu iya ƙirƙirar tsarin hankali.
Hakanan za mu bi ta tsarin da TensorFlow ke bayarwa don ƙirƙirar samfuran AI. Don haka bari mu fara!
Takaitaccen Gabatarwa zuwa TensorFlow
TensorFlow na Google shine tushen bude-bude injin inji kunshin software. Ya haɗa da kayan aikin horo da turawa samfurin koyo na inji akan dandamali da yawa. da na'urori, kazalika da goyan bayan zurfafa ilmantarwa da neural networks.
TensorFlow yana ba masu haɓaka damar ƙirƙirar samfura don aikace-aikace iri-iri. Wannan ya haɗa da gano hoto da sauti, sarrafa harshe na halitta, da hangen nesa na kwamfuta. Kayan aiki ne mai ƙarfi kuma mai daidaitawa tare da yaɗuwar tallafin al'umma.
Don shigar da TensorFlow akan kwamfutarka zaku iya rubuta wannan a cikin taga umarnin ku:
pip install tensorflow
Ta yaya Model AI ke Aiki?
Samfuran AI tsarin kwamfuta ne. Don haka, ana nufin su yi ayyukan da za su buƙaci hankalin ɗan adam. Gane hoto da magana da yanke shawara misalai ne na irin waɗannan ayyuka. Ana haɓaka samfuran AI akan manyan bayanan bayanai.
Suna amfani da dabarun koyon injin don samar da tsinkaya da aiwatar da ayyuka. Suna da amfani da yawa, gami da motoci masu tuƙi, mataimakan kai, da binciken likita.
Don haka, menene shahararrun samfuran TensorFlow AI?
Sake saiti
ResNet, ko Residual Network, wani nau'i ne na juyin juya hali neural network. Muna amfani da shi don rarraba hoto da kuma gano abu. Masu bincike na Microsoft ne suka haɓaka shi a cikin 2015. Hakanan, an bambanta shi ta hanyar amfani da ragowar haɗin gwiwa.
Waɗannan haɗin gwiwar suna ba da damar cibiyar sadarwa ta koya cikin nasara. Don haka, yana yiwuwa ta hanyar ba da damar bayanai su gudana cikin 'yanci tsakanin yadudduka.
Ana iya aiwatar da ResNet a cikin TensorFlow ta amfani da Keras API. Yana ba da babban matakin, mai sauƙin amfani don ƙirƙira da horar da cibiyoyin sadarwar jijiyoyi.
Ana shigar da ResNet
Bayan shigar da TensorFlow, zaku iya amfani da Keras API don ƙirƙirar ƙirar ResNet. TensorFlow ya haɗa da Keras API, don haka ba kwa buƙatar shigar da shi daban-daban.
Kuna iya shigo da samfurin ResNet daga tensorflow.keras.applications. Kuma, zaku iya zaɓar sigar ResNet don amfani, misali:
from tensorflow.keras.applications import ResNet50
Hakanan zaka iya amfani da lambar mai zuwa don loda ma'aunin nauyi da aka riga aka horar don ResNet:
model = ResNet50(weights='imagenet')
Ta zaɓin kadarorin sun haɗa da_top=Ƙarya, kuna iya amfani da samfurin don ƙarin horo ko daidaita saitunan bayananku na al'ada.
model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
Wuraren Amfani da ResNet
Ana iya amfani da ResNet a cikin rarraba hoto. Don haka, zaku iya rarraba hotuna zuwa rukuni da yawa. Da farko, kuna buƙatar horar da ƙirar ResNet akan babban tarin bayanai na hotuna masu lakabi. Sa'an nan, ResNet na iya yin tsinkaya ajin hotunan da ba a gani a baya.
Ana iya amfani da ResNet don ayyukan gano abubuwa kamar gano abubuwa a cikin hotuna. Za mu iya yin haka ta farko da horar da samfurin ResNet akan tarin hotuna da aka yi wa lakabi da akwatunan daure abu. Bayan haka, zamu iya amfani da samfurin da aka koya don gane abubuwa a cikin sabbin hotuna.
Hakanan zamu iya amfani da ResNet don ayyukan rarrabuwar kawuna. Don haka, za mu iya sanya lakabin ma’ana ga kowane pixel a cikin hoto.
kafuwarta
Ƙirƙirar ƙirar ilmantarwa mai zurfi ce mai iya gane abubuwa a cikin hotuna. Google ya sanar da shi a cikin 2014, kuma yana nazarin hotuna masu girma dabam ta amfani da yadudduka da yawa. Tare da Ƙaddamarwa, ƙirar ku na iya fahimtar hoton daidai.
TensorFlow kayan aiki ne mai ƙarfi don ƙirƙira da gudanar da samfuran Inception. Yana ba da babban matakin haɗin kai da mai amfani don horar da cibiyoyin sadarwar jijiyoyi. Don haka, Inception kyakkyawan tsari ne mai sauƙi don nema ga masu haɓakawa.
Shigarwa Ƙaddamarwa
Kuna iya shigar da Inception ta buga wannan layin lambar.
from tensorflow.keras.applications import InceptionV3
Wuraren Amfani
Hakanan ana iya amfani da samfurin Ƙaddamarwa don fitar da fasali a ciki zurfin ilmantarwa samfura kamar Generative Adversarial Networks (GANs) da Autoencoders.
Za a iya daidaita samfurin Ƙaddamarwa don gano takamaiman halaye. Har ila yau, ƙila za mu iya gano wasu cututtuka a aikace-aikacen hoto na likita kamar X-ray, CT, ko MRI.
Za a iya daidaita samfurin Ƙaddamarwa don bincika ingancin hoto. Za mu iya tantance ko hoton yana da duhu ko kintsattse.
Ana iya amfani da ƙaddamarwa don ayyukan nazarin bidiyo kamar bin diddigin abu da gano ayyuka.
BERT
BERT (Wakilan Encoder na Bidirectional daga Masu Canjawa) samfurin hanyar sadarwa ne na Google wanda aka riga aka horar da shi. Za mu iya amfani da shi don ayyuka daban-daban na sarrafa harshe na halitta. Waɗannan ayyuka na iya bambanta daga rarrabuwar rubutu zuwa amsa tambayoyi.
BERT an gina shi akan gine-ginen transfoma. Don haka, zaku iya sarrafa ɗimbin shigarwar rubutu yayin fahimtar haɗin kalmomi.
BERT samfurin da aka riga aka horar da shi wanda zaku iya haɗawa cikin aikace-aikacen TensorFlow.
TensorFlow ya haɗa da samfurin BERT da aka riga aka horar da kuma tarin kayan aiki don daidaitawa da amfani da BERT zuwa ayyuka iri-iri. Don haka, zaku iya haɗa ƙaƙƙarfan iyawar sarrafa harshe na BERT cikin sauƙi.
Shigar da BERT
Yin amfani da mai sarrafa fakitin pip, zaku iya shigar da BERT a cikin TensorFlow:
pip install tensorflow-gpu==2.2.0 # This installs TensorFlow with GPU support
pip install transformers==3.0.0 # This installs the transformers library, which includes BERT
Ana iya shigar da nau'in CPU na TensorFlow cikin sauƙi ta hanyar maye gurbin tensorflow-gpu tare da tensorflow.
Bayan shigar da ɗakin karatu, zaku iya shigo da samfurin BERT kuma kuyi amfani da shi don ayyukan NLP daban-daban. Anan akwai lambar samfurin don daidaita ƙirar BERT akan matsalar rarraba rubutu, misali:
from transformers import BertForSequenceClassification
# Load the pre-trained BERT model
model = BertForSequenceClassification.from_pretrained("bert-base-uncased")
# Fine-tune the model on your text classification task
model.fit(training_data, labels)
# Make predictions on new data
predictions = model.predict(test_data)
Yankunan Amfani na BERT
Kuna iya yin ayyukan rarraba rubutu. Misali, yana yiwuwa a cimma tantance tunani, Rarraba batutuwa, da gano spam.
BERT yana da Amincewa da Mahaɗan (NER) fasali. Don haka, zaku iya ganewa da yiwa ƙungiyoyi lakabi a cikin rubutu kamar mutane da ƙungiyoyi.
Ana iya amfani da shi don amsa tambayoyin dangane da wani mahallin musamman, kamar a cikin injin bincike ko aikace-aikacen chatbot.
BERT na iya zama da amfani ga Fassara Harshe don ƙara daidaiton fassarar inji.
Ana iya amfani da BERT don taƙaita rubutu. Don haka, yana iya bayar da taƙaitaccen bayani mai amfani na dogayen takaddun rubutu.
DeepVoice
Binciken Baidu ya kirkiro DeepVoice, a rubutu-zuwa-magana samfurin kira.
An halicce shi tare da tsarin TensorFlow kuma an horar da shi akan babban tarin bayanan murya.
DeepVoice yana haifar da murya daga shigar da rubutu. DeepVoice yana ba da damar ta amfani da dabarun koyo mai zurfi. Samfurin tushen hanyar sadarwa ne na jijiya.
Don haka, yana nazarin bayanan shigarwa kuma yana haifar da magana ta amfani da ɗimbin adadin yadudduka na nodes.
Sanya DeepVoice
!pip install deepvoice
A madadin;
# Clone the DeepVoice repository
!git clone https://github.com/r9y9/DeepVoice3_pytorch.git
%cd DeepVoice3_pytorch
!pip install -r requirements.txt
Abubuwan Amfani na DeepVoice
Kuna iya amfani da DeepVoice don samar da magana don mataimaka na sirri kamar Amazon Alexa da Google Assistant.
Hakanan, ana iya amfani da DeepVoice don samar da magana don na'urori masu kunna murya kamar lasifika masu wayo da tsarin sarrafa kansa na gida.
DeepVoice na iya ƙirƙirar murya don aikace-aikacen maganin magana. Zai iya taimaka wa marasa lafiya da matsalolin magana don inganta maganganun su.
Ana iya amfani da DeepVoice don ƙirƙirar magana don abubuwan ilimantarwa kamar littattafan mai jiwuwa da ƙa'idodin koyon harshe.
Leave a Reply