АНАЛИЗ ВОЗМОЖНОСТЕЙ GOOGLE COLAB

Волокитина Татьяна Сергеевна
Юго-Западный государственный университет

Аннотация
Статья посвящена описанию возможности Google Colab для изучения технологий машинного обучения и нейронных сетей. Автор предоставляет теоретическую справку и анализ возможностей Google Colab. В данной работе используется файл GitHub для примера работы сортировки изображений нейросетью.

ANALYSIS OF GOOGLE COLAB POSSIBILITIES

Volokitina Tatiana Sergeevna
South-West State University

Abstract
The article is devoted to the description of Google Colab possibilities for studying machine learning technologies and neural networks. The author provides theoretical background and analysis of Google Colab features. This paper uses GitHub file for an example of neural network image sorting performance.

Keywords: FastAI, GitHub, Google Colab, machine learning, neural network


Рубрика: 05.00.00 ТЕХНИЧЕСКИЕ НАУКИ

Библиографическая ссылка на статью:
Волокитина Т.С. Analysis of Google Colab possibilities // Современные научные исследования и инновации. 2021. № 2 [Электронный ресурс]. URL: https://web.snauka.ru/issues/2021/02/94760 (дата обращения: 29.03.2024).

In today’s world, neural networks are used everywhere to find logic and predictions from the analysis of data fed into the network. A system of neurons is a self-learning system that acts similarly to the human brain. The unit of the neural network is the neuron, which was prototyped as the brain unit of the same name. Figure 1 shows the scheme of the neuron.


Figure 1 – Schematic diagram of a neural network unit

It can be seen that the components of a neural network unit are the synapses that provide the link between the inputs and the core; while the neuron core is used to directly analyse the inputs; and the axon provides the link between the core and the next layer of the neural network or output.

In this paper an example of using Google Cloud for machine learning has been considered. For its implementation we will use GitHub file with neural network. Similar to FastAI, we will create a neural network that will classify images from our own dataset.

First we need to prepare images for classification, we will classify black bears and grizzly. We have to delete all the bad or damaged images from Google Images. You can do this with a loop:

for c in classes:

print(c)

verify_images(path/c, delete=True, max_size=500)

Let’s create dataset, namely data set.

Dataset is a set of images for the classification of black bears (black) and grizzly, which will be input (Figure 2).


Figure 2 – Classification dataset

After the dataset has been successfully created, it is necessary to begin the training of the neural network. To do this we will use the following function:

learn = create_cnn(data, models.resnet34, metrics=error_rate)

We will get a table with the data on training of neural networks (Fig. 3):


Figure 3 – Training.

After the successful training let us see the interpretation of the results (fig. 4).


Figure 4 – Interpretation of the classification

As you can see in the picture, 41 images of black bears and 31 images of grizzly bears were correctly recognized, but a few errors were found which we will examine more closely (figure 5).


Figure 5 – Erroneous images

You can see from the graph that the pictures cannot be interpreted accurately, which affects the result.

Let us test the trained neural network. Let’s take the image of grizzly and check the classification accuracy (figure 6):


Figure 6 – Image for the test

Let’s get the output from the neural network:

Category grizzly, which means the classification occurred correctly.

As a result of this work, we have described the possibilities of Google Colab for studying the machine learning technologies and neural networks and presented an example of classification of images taken from Google Image.


References
  1. Barsky, A.B. Logical neural networks: Tutorial / A.B. Barsky. – Moscow: Binom, 2013. – 352 c.
  2. Galushkin, A.I. Neural networks: basics of theory / A.I. Galushkin. – M.: GLT, 2012. – 496 c.
  3. Callan, R. Neural networks: A quick reference guide / R. Callan. – Moscow: Williams I.D., 2017. – 288 c.
  4. Redko, V.G. Evolution, neural networks, intelligence: Models and concepts of evolutionary cybernetics / V.G. Redko. – Moscow: Lenand, 2019. – 224 c.


Количество просмотров публикации: Please wait

Все статьи автора «Волокитина Татьяна Сергеевна»


© Если вы обнаружили нарушение авторских или смежных прав, пожалуйста, незамедлительно сообщите нам об этом по электронной почте или через форму обратной связи.

Связь с автором (комментарии/рецензии к статье)

Оставить комментарий

Вы должны авторизоваться, чтобы оставить комментарий.

Если Вы еще не зарегистрированы на сайте, то Вам необходимо зарегистрироваться:
  • Регистрация