AI, practical course. Foreword

Intel's blog begins publishing a Russian translation of a large series of articles from the specialized Intel AI Academy resource. The purpose of this publication is to present various approaches to artificial intelligence and various ways of its application. The first post of the series will be in some way a preface: here you will find an introductory part from the authors of the course, as well as a complete list of articles in English and (as published) in Russian.

We hope that our course will be useful for you.

The revolutionary advances in artificial intelligence (AI), machine and in-depth learning transform our familiar understanding of software. Both well-known technology giants and newly minted startups use AI capabilities to solve new problems, including unmanned vehicles, virtual personal assistants, the development of new drugs and forecast trends in financial markets. And although the list of today's uses for AI is extensive and varied, it barely outlines what awaits us in the future.
')
As part of this curriculum, experts in the field of AI, machine learning and in-depth training will introduce you to the tools, infrastructure, and techniques of AI, demonstrating the process of creating an application using the rich capabilities of AI.

Artificial intelligence in action

Using the example of an automatic video editing application, you will learn about two main difficulties of AI: image classification and sequence prediction. We will talk about how to use convolutional neural networks to classify images, that is, how to automatically recognize emotions in images from the original set. Next, you will learn how to use recurrent neural networks to synthesize a musical melody based on algorithms to accompany emotions in images. The result should be a finished video with a soundtrack created by computer means.



Throughout the entire series of educational materials, we will demonstrate various concepts of artificial intelligence and tell you about the architecture of Intel, which is the basis of deep neural networks. We will show you how to optimize the process of writing code for AI applications using modern Intel technologies, including:

• Intel Data Analytics Acceleration Library (Intel DAAL) ;
• Intel Math Kernel Library (Intel MKL);
• Intel Distribution for Caffe ;
• library for deep learning TensorFlow .

We will look at the main stages of the process of developing artificial intelligence: formulating a concept, creating a team, collecting and storing data, developing and evaluating models, and deploying. We will also analyze all the important points of decision making by comparing different algorithms and techniques of AI, as well as options for data centers and cloud infrastructure, as you would have done while working on your application.

Required skills

Composing these teaching materials, we proceeded from the fact that students will have an average level of knowledge of the Python programming language, a basic level of knowledge of linear algebra, statistics and probability theory, as well as some familiarity with GitHub. But even if you do not have all this knowledge, you can enroll in the course and follow how we share the source code and configurations suitable for cloning, then to apply them to your own AI applications. Non-technical specialists can get an idea and information on how AI applications are developed. We will even introduce you to the Docker containers, the Keras neural network application interface , the TensorFlow software for machine intelligence, and the Caffe depth learning methodology.

The target audience

These training materials can be used by everyone, but when writing them we focused mainly on the following specialists:

• Experts in data analysis and processing , who would like to learn about the Intel software suite for in-depth training and artificial intelligence, as well as see how Intel's AI technology behaves in real-world applications.
• Software developers who would like to quickly learn machine and in-depth training, having completed a practical course of study.
• Product managers and data center managers who would like to get acquainted with the latest AI technologies from Intel and learn from professionals how to organize an AI project in order to increase the chances of success.

Summary of Tutorials

The whole series of educational articles on the practical application of artificial intelligence is divided into five stages. At each stage we will consider several main topics:

• The idea . Start your project by defining a common set of goals, data sources, technological constraints, team members and formulating a strategic plan.
• Technology . Explore different approaches, learn how to choose the right approach according to the needs and long-term goals of your project, and find out what are the options for data centers and cloud environments to perform your computing tasks.
• The data . Learn how to properly collect, analyze and pre-process data, as well as add comments to it.
• Model Learn how to build an experimental infrastructure, select and train a model, and iterate through it until you are satisfied with the result.
• Application Deploy the model in a production environment using a user-friendly interface (for intelligent applications that interact with users).

Course contents

Original articles in English

1. Create Applications with Powerful AI Capabilities
2. Ideation
3. The Anatomy of an AI Team
4. Project Planning
5. Select a Deep Learning Framework
6. Select an AI Computing Infrastructure
7. Amazon Mechanical Turk
8. Crowdsourcing Word Selection for Image Search
9. Data Annotation Techniques
10. Set Up a Portable Experimental Environment for Deep Learning with Docker
11. Image Dataset Search
12. Image Data Collection
13. Image Data Exploration
14. Image Data Preprocessing and Augmentation
15. Overview of Convolutional Neural Networks for Image Classification
16. Modern Deep Neural Network Architectures for Image Classification
17. Emotion Recognition from Images Baseline Model
18. Emotion Recognition from Images Model Tuning and Hyperparameters
19. Music Dataset Search
20. Music Data Collection and Exploration
21. Emotion-Based Music Transformation
22. Deep Learning for Music Generation 1 — Choosing a Model and Data Preprocessing
23. Deep Learning for Music Generation 2 — Implementing the Model

Translated articles (as published)

1. Foreword
2. Project planning
3. Comparison of deep learning software
4. Collection and study of images
5. Preprocessing and supplementing data with images
6. Neural network overview for image classification
7. Modern deep neural network architectures for image classification
8. The basic model of recognition of emotions in images
9. Configure the model and hyperparameters to recognize emotions in images
10. Emotion Based Musical Transformation
11. Deep learning to generate music