- Download 0
- File Size 7.72 MB
- File Count 1
- Create Date October 12, 2020
- Last Updated May 23, 2021
Python Tutorial | Azure Tutorial | Deep Learning Tutorial | Deep Learning with Azure
Deep Learning with Azure Building and Deploying Artificial Intelligence Solutions on the Microsoft AI Platform
This book spans topics such as general techniques and frameworks for deep learning, starter guides for several approaches in deep learning, and tools, services, and infrastructure for developing and deploying AI solutions using the Microsoft AI platform. This book is primarily targeted to data scientists who are familiar with basic machine learning techniques but have not used deep learning techniques or who are not familiar with the Microsoft AI platform. A secondary audience is developers who aim for an introduction to AI and getting started with the Microsoft AI platform.
It is recommended that you have a basic understanding of Python and machine learning before reading this book. It is also useful to have access to an Azure subscription to follow along with the code examples and get the most benefit from the material, although it is not required to read the book.
How This Book Is Organized
In Part I of the book, we introduce the basic concepts of AI and the role Microsoft has related to AI solutions. Building on decades of research and technological innovations, Microsoft now provides services and infrastructure to enable others who want to build intelligent applications with the Microsoft AI platform built on top of the Azure cloud computing platform.
We introduce machine learning and deep learning in the context of AI and explain why these have become especially popular in the last few years for many different business applications. We outline example use cases utilizing AI, especially employing deep learning techniques, which span from several verticals such as manufacturing, health care, and utilities.
In the first part of the book, we also give an overview of deep learning, including common types of networks and trends in the field. We also discuss limitations of deep learning and go over how to get started.
In Part II, we give a more in-depth overview of the Microsoft AI platform. For data scientists and developers getting started using AI in their applications, there are a range of solutions that are useful in different situations. The specific services and solutions will continue to evolve over time, but two main categories of solutions are available.
The first category is custom solutions built on the Microsoft Azure AI platform. Chapter 4, “Microsoft AI Platform,” discusses the services and infrastructure on the Microsoft AI platform that allow one to build custom solutions, especially Azure Machine Learning services for accelerating the life cycle of developing machine learning applications as well as surrounding services such as Batch AI training and infrastructure such as the Deep Learning Virtual Machine.
The second category is Microsoft’s Cognitive Services, which are pre-trained models that are available as a REST application programming interface (API). In other words, the models are already built on a set of data and users can use the pre-trained model. Some of these are ready to use without any customization. For example, there is a text analytics service that allows one to submit the text and get a sentiment score for how positive or negative the text is. This type of service could be useful in analyzing product feedback, for example. Other Cognitive Services are customizable, where you can bring your own data to customize the model. These services are covered in more detail in Chapter 5, “Cognitive Services and Custom Vision.”
In Part III, we cover three common types of deep learning models—convolutional neural networks, recurrent neural networks, and generative adversarial networks—that are useful to understand in building out custom AI solutions. Each chapter includes links to code samples for understanding the type of network and how one can build such a network using the Microsoft AI platform.
In the final part of the book, Part IV, we consider architecture choices for building AI solutions using the Microsoft AI platform along with sample code. Specifically, Chapter 9, “Training AI Models,” covers options for training neural networks such as Batch AI service and DL workspace.
Chapter 10, “Operationalizing AI Models,” covers deployment options for scoring neural networks such as Azure Kubernetes Service for serving real-time models as well as Spark using the open source library MMLSpark from Microsoft.