Neural networks. Computer vision and reading (NLP). — rate 31990 rub. from Specialist, training 24 academic hours, date: December 11, 2023.
Miscellaneous / / December 03, 2023
Neural networks - firmly established modern content processing technology. Today, many computer IT corporations use this technology to create computer robots and chat bots. The most famous of them Alexa (Amazon), Siri (Apple), Alice (Yandex), O'key Google, Google Translate (Google) were created using this technology.
This course will examine a number of neural networks implemented in Python using the Tensorflow library, namely PyTorch, developed in 2017. These algorithms form the basis for solving problems in computer vision and reading, but do not exhaust it, since this area is constantly developing and improving.
- interact with tensors in Python
- get acquainted with the basics of PyTorch
- deepen your knowledge of Python
- get acquainted with image processing using neural networks and Python
- become familiar with speech and text processing
Teacher of Python courses for machine learning. Vladimir Gennadievich is an experienced practitioner, candidate of physical and mathematical sciences and active researcher.
In his work, he uses methods of machine learning and automation of data collection using the programming languages Python, R, C++, Verilog.
Vladimir Gennadievich is a member of the Research Gate community of researchers and constantly monitors how programming is used in science and modern developments. He shares with his listeners know-how and current techniques that will help make their projects better and world-class.
Vladimir Gennadievich published 56 articles in such publications as Physical Review B, Physica E, “Journal of Experimental and Theoretical Physics”, “Physics and Technology of Semiconductors”. Vladimir Gennadievich not only participates in the development of science and shares his achievements with colleagues, but also successfully uses them in practice:
Vladimir Gennadievich, as a teacher-scientist, puts the development and application of new technologies in first place. In learning, including machine learning, the main thing for him is to penetrate into the essence of phenomena, to understand all the processes, and not to memorize the rules, code or syntax of technical means. His credo is practice and deep immersion in work!
Practical teacher with 25 years of experience in the field of information technology. Expert in Full-Stack development of web systems using (MySQL, PHP/Python, nginx, HTML5), data analysis and visualization using Python (Pandas, SKLearn, Keras), development...
Practical teacher with 25 years of experience in the field of information technology. Expert in Full-Stack development of web systems using (MySQL, PHP/Python, nginx, HTML5), data analysis and visualization using Python (Pandas, SKLearn, Keras), development of data exchange interfaces between systems using REST, SOAP, EDIFACT technologies, administering web servers on Debian GNU Linux (php/nginx/mariadb), creating technical and user documentation (in Russian and English languages).
I went all the way from a line developer to the IT director of my own company. Over 25 years, he has created about 20 corporate information systems/databases, more than 50 prototypes, 30 websites of varying sizes and content. Worked on large projects for companies such as Maersk, Toyota, Nissan, Rossiya-on-Line, Glasnet. For 5 years he has been among the TOP 10 developers in the Russian Federation on phpClasses.org.
Module 1. Introduction to Pytorch and tensors (4 ac. h.)
- Introduction to the course
- Introduction to neural networks
- What is PyTorch?
- Why use tensors?
- Technical requirements
- Cloud capabilities
- What are tensors
- Operations with tensors
- Workshop on the topic
Module 2. Image classification (4 ac. h.)
- Tools for loading and processing data in PyTorch
- Creating a training data set
- Creation of validation and test data set
- Neural network as tensors
- Activation function
- Network creation
- Loss function
- Optimization
- Workshop, implementation on GPU
Module 3. Convolutional neural networks (6 ac. h.)
- Building a simple convolutional neural network in PyTorch
- Combining layers in a network (Pooling)
- Neural network regularization (Dropout)
- Use of trained neural networks
- Study of the structure of the neural network
- Batch normalization (Batchnorm)
- Workshop on the topic
Module 4. Use and transfer of trained models (5 ac. h.)
- Using ResNet
- Selection by learning speed
- Learning Rate Gradient
- Data expansion for retraining
- Using Torchvision converters
- Color and lambda converters
- Custom converters
- Ensembles
- Workshop on the topic
Module 5. Text classification (5 ac. h.)
- Recurrent neural networks
- Neural networks with memory
- Torchtext Library