微软认知服务与TensorFlow实战：AI解决方案指南

需积分: 10 190 浏览量更新于2024-07-17 1 收藏 23.94MB PDF 举报

"Apress.Cognitive.Computing.Recipes"是一本专注于人工智能和机器学习实践的实用指南，它通过Microsoft Cognitive Services和TensorFlow等工具，帮助读者解决实际问题。作者Adnan Masood和Adnan Hashmi采用问题解决的方法论，确保读者能够理解和应用深度学习和机器学习技术于日常生活中的开发工作。本书的核心内容围绕以下几个方面展开： 1. 入门与技术概述：首先，它提供了一个现代技术领域的概览，帮助读者理解AI和机器学习在当今技术栈中的位置，以及它们如何与其他技术如自然语言处理、计算机视觉、文本分析、语音识别和机器人过程自动化相结合。 2. Microsoft Cognitive Services的应用：书中特别强调了Microsoft Cognitive Services在构建生产级解决方案中的作用。读者将学习如何利用这些服务API来开发具有竞争力的应用程序，如智能助手或实现高效的业务流程。 3. 深度学习技术：通过TensorFlow和Microsoft Cognitive Toolkit (CNTK)的结合，本书展示了如何在实际项目中运用深度学习技术，包括模型训练、部署和优化的过程。 4. 解决企业级挑战：书中提供了针对自然语言处理和计算机视觉等领域的具体解决方案，帮助企业解决实际业务问题，提升效率和决策能力。 5. 机器学习生命周期：书中的教程覆盖了机器学习项目的全生命周期，从问题定义、模型设计、开发到大规模部署，确保读者掌握完整的开发流程。这本书适合的读者是希望通过实践学习和解决问题的软件工程师和企业架构师，他们可以通过这本书深入了解和掌握机器学习和深度学习技术，并将其应用于实际业务场景中，推动组织的数字化转型。 "Apress.Cognitive.Computing.Recipes"不仅提供了一套丰富的工具和技术，还为读者提供了一个实际操作的平台，让他们能够在探索AI的无限可能时，稳步前进并取得成功。

xviii

Machine learning does not exist without data. Whether in bulk or in streams, models

are trained, evaluated, and improved through data. Machine learning lets us extract much

greater value out of data of all shapes and sizes. Machine learning can even be used to

enrich data. Consider the most recent code that you wrote dealing with user input. Odds

are, it dealt with numerical values, or relatively short string values. Modern programming

languages are incredibly powerful and efficient when dealing with those data types.

Expand that out a little bit, to images, video, sounds, or large volumes of text. What data

types exist to reason on these, not just manipulate them? Machine learning enriches data

by letting us address a larger variety of data, and by transforming it into things that can be

reasoned over in code. Cognitive APIs, like Microsoft’s Cognitive Services make it easy to

take an image and break it into component parts. Want to know what’s in the image, how

many people are in it, are they happy? All of that is available to you with a simple HTTP

method. The output of that easily goes into our programs to make a decision, such as

adjusting a thermostat based on the number of people in a room.

Devices are becoming smarter, and in many cases, more connected. Machine

learning models, built on data observed from those devices, enable better understanding

of the device and the environment around it. This enables more efficient devices to

be built, future designs to be influenced, but more importantly, can be used to predict

failures or identify anomalies. The “digital exhaust” from these devices is valuable, not

only for training new models, but also for being able to provide a mechanism to evaluate

the impact and outcome of models currently deployed. This stream of exhaust is key to

creating the virtuous cycle of model development, improvement, and improvements in

outcomes.

One of the most inspiring examples of this is Microsoft’s AI for Earth initiative, which

looks to provide grant funding to organizations leveraging AI to advance sustainability.

I’ve had the opportunity to interact with a number of the grantees, and the stories

of how they are transforming the way we consume, conserve, and manage natural

resources remind us of the power of software as a force for good. Every industry is being

transformed, and that is being powered by machine learning.

The other key aspect of “Why now,” is the cloud. The emergence of plentiful,

and powerful, compute capacity in the cloud, along with the hardware and software

innovations in the GPU space has helped to foster a massive wave of innovation, with

much of that coming in the area of deep learning. The fundamentals behind deep

learning are not new, neural networks modeling the functionality of the neurons in the

brain originated in the middle of the last century. Three things needed to come together

foreword

xxi

Preface

The hype around disruptive new technologies travels from the peak of inflated

expectations to the valley of disillusionment until it hits its stride in the plateau of

productivity. This journey is often counterproductive, and AI and machine learning are

not an exception to this rule. The jury is still out on why it is called deep-learning hype

and not backpropaganda, but I digress. It is safe to say that the ubiquitous nature of

machine learning–based applications enables today’s commerce and innovation. As one

discerns what is possible with AI for one’s business, the concrete use cases are no longer

limited by lack of vision and instead become quite evident. This figurative leap of faith

requires a future-oriented growth mindset and thought leadership. We quickly see the

key concerns from C-suite start with Why AI, quickly followed by How and then When!

This book you are holding is written with the motive of exploring the art of the

possible with its readers. This manuscript is by no means perfect; however, it intends

to give you tools and an understanding of the contemporary AI landscape. Intelligent

correlation using AI and machine learning helps companies achieve richer search

experiences; where and how to use this to accomplish real productivity gains requires

you, the domain expert, to connect your business use case with the algorithms and

models. AI data insights and discovery are transforming customer service into a science

where previously there was only art. I implore you to explore novel use cases across the

spectrum, such as bias-free product-recommendation engines, detection of potential

predatory tactics with natural language annotations, contract analysis to augment and

accelerate the discovery process, inventory monitoring and analysis using computer

vision, and digital assistants that can comprehend, summarize, and answer queries on

diverse enterprise corpora; the possibilities are bounded only by your imagination.

Enterprise-grade AI search is helping enterprises cut through the fog of uncertainty,

helping in product development, and extracting real business value by intelligent

process compression.

Let’s embark on the AI-first journey of scholastic gradient descent.

Adnan Masood, PhD

Chief Architect AI & Machine Learning—Microsoft MVP (Artificial Intelligence); visiting

scholar, Department of Computer Engineering, Stanford University.

剩余436页未读，继续阅读

ALo54

粉丝: 5
资源: 111

微软认知服务与TensorFlow实战：AI解决方案指南

最新资源