Artificial Intelligence By Example

Acquire advanced AI, machine learning, and deep learning design skills, 2nd Edition

Denis Rothman

• Publisher: Packt Publishing
• Year: 2020
• eBook ISBN: 9781839212819
• Edition: 2
• Topic: Informatica
• Subtopic: Intelligenza artificiale (IA) e semantica

Understand the fundamentals and develop your own AI solutions in this updated edition packed with many new examples

Key Features

• AI-based examples to guide you in designing and implementing machine intelligence
• Build machine intelligence from scratch using artificial intelligence examples
• Develop machine intelligence from scratch using real artificial intelligence

Book Description

AI has the potential to replicate humans in every field. Artificial Intelligence By Example, Second Edition serves as a starting point for you to understand how AI is built, with the help of intriguing and exciting examples.

This book will make you an adaptive thinker and help you apply concepts to real-world scenarios. Using some of the most interesting AI examples, right from computer programs such as a simple chess engine to cognitive chatbots, you will learn how to tackle the machine you are competing with. You will study some of the most advanced machine learning models, understand how to apply AI to blockchain and Internet of Things (IoT), and develop emotional quotient in chatbots using neural networks such as recurrent neural networks (RNNs) and convolutional neural networks (CNNs).

This edition also has new examples for hybrid neural networks, combining reinforcement learning (RL) and deep learning (DL), chained algorithms, combining unsupervised learning with decision trees, random forests, combining DL and genetic algorithms, conversational user interfaces (CUI) for chatbots, neuromorphic computing, and quantum computing.

By the end of this book, you will understand the fundamentals of AI and have worked through a number of examples that will help you develop your AI solutions.

What you will learn

• Apply k-nearest neighbors (KNN) to language translations and explore the opportunities in Google Translate
• Understand chained algorithms combining unsupervised learning withdecision trees
• Solve the XOR problem with feedforward neural networks (FNN) and buildits architecture to represent a data flow graph
• Learn about meta learning models withhybrid neural networks
• Create a chatbot and optimize its emotional intelligence deficiencies withtools such as Small Talk and data logging
• Building conversational user interfaces (CUI) for chatbots
• Writing genetic algorithms that optimizedeep learning neural networks
• Build quantum computing circuits

Who this book is for

Developers and those interested in AI, who want to understand the fundamentals of Artificial Intelligence and implement them practically. Prior experience with Python programming and statistical knowledge is essential to make the most out of this book.

11

Combining Reinforcement Learning and Deep Learning

Amazon is one of the world's leading e-retailers, with sales over US$ 250 billion. Amazon's e-store sales exceed all of its other activities, such as AWS subscription services (premium, for example), retail third-party seller services, and physical stores.

This chapter focuses on apparel production activity, one of Amazon's markets for which the company recently registered a patent. Prime Wardrobe even offers a try-and-easy-return service. This new activity requires planning and scheduling. Amazon took the matter seriously and registered a patent for an apparel manufacturing system to control the production process of its apparel products.

Artificial intelligence already plays a role in automatic planning and scheduling in the apparel business, from customer orders through to delivery.

Google has successfully combined deep learning and reinforcement learning (Q-learning) in a deep Q-network (DQN), a system that can beat humans at video games and other tasks. Google's AlphaGo DQN has obtained impressive results.

In this chapter, we go beyond merely describing Amazon's process or Google's process in particular. I added on my real-life implementations of what we will explore and build in Python from scratch. As such, we will combine the ideas of several systems (Amazon, Google, and my implementations).

We will add an innovation for the apparel manufacturing industry by adding a conceptual representation learning metamodel (CRLMM) to reinforcement learning.

We will go from scratch to a prototype that could be implemented on-site, the foundations are being established for further applications in the coming chapters.

The following topics will be covered in this chapter:

• Planning and scheduling today and tomorrow
• Further generalization of the CRLMM described in Chapter 10, Conceptual Representation Learning, applied to an apparel production process
• Feeding the CRLMM convolutional neural network (CNN) with a simulation of frames coming from a webcam on a production line
• Introducing an optimizer that will use weights applied to production stations to input a reward matrix to a Markov decision process (MDP), which will then update the weights
• Building a program that will run continuously (no beginning, no end) on a production line using all the three components mentioned previously

We'll begin by talking about planning and scheduling today and tomorrow. The market is slowing moving from preplanned processes to real-time processes. Let's see how.

Planning and scheduling today and tomorrow

When Amazon decided to launch Prime Wardrobe, it brought a new service to its customers, enabling them to order, try out, and purchase clothing, shoes, and other accessories. The customer can establish a purchase plan. A purchase plan is a list of tasks to be carried out in a given time. An example of a purchase plan could be:

• Filling a box with clothing
• Trying on the clothing at home
• Returning the clothing if it does not fit
• Purchasing the items that are kept

Once the customer agrees to follow this plan, the time sequence becomes crucial:

• First, a box must be chosen.
• Then, there is a delivery period.
• Then, there is a trial period (you cannot try the products forever). During this period, the customer can choose not to purchase anything.
• Finally, the customer confirms the purchase.

Irrespective of whether Amazon Prime Wardrobe will remain a service in years to come, the precedent is set; just as physical book stores disappear every year, shopping for clothing online will continue to expand and occupy more of the market.

On top of that distribution, corporations will continue to expand their production sites to become manufacturing-distributing giants. Warehouses will progressively replace many stores as per the warehouse examples provided in some of the chapters in this book.

Supply chain management (SCM), combined with APS, has become a necessity. SCM-APS constraints vary constantly on a global market (depending on the manufacturer). APS stands for advanced planning and scheduling or automated planning and scheduling. We will explore the difference between these two concepts in a following section, A real-time manufacturing revolution. We will go beyond Amazon's approach to the subject since delivering in real-time has become a constraint for all of the actors on the market.

The pressure of the market has encouraged Amazon to produce its own clothing. Amazon has produced its in-house fashion labels with apparel and accessories sold worldwide.

To prove that it means business, Amazon registered several patents, including one for a blended reality mirror, an apparel manufacturing system, and more. With a blended reality mirror, a person can visualize how the clothing will fit.

In the following sections, we will explore the planning and scheduling side of Amazon's apparel in-house manufacturing plans, which will have the same effect on apparel factories as it did on physical book stores and all types of shops. Many new jobs will emerge, such as hundreds of thousands of jobs in the artificial intelligence business, on websites, and in marketing and SCM. Many jobs will disappear as well. The improvements...

Table of contents

• Preface
• Getting Started with Next-Generation Artificial Intelligence through Reinforcement Learning
• Building a Reward Matrix – Designing Your Datasets
• Machine Intelligence – Evaluation Functions and Numerical Convergence
• Optimizing Your Solutions with K-Means Clustering
• How to Use Decision Trees to Enhance K-Means Clustering
• Innovating AI with Google Translate
• Optimizing Blockchains with Naive Bayes
• Solving the XOR Problem with a Feedforward Neural Network
• Abstract Image Classification with Convolutional Neural Networks (CNNs)
• Conceptual Representation Learning
• Combining Reinforcement Learning and Deep Learning
• AI and the Internet of Things (IoT)
• Visualizing Networks with TensorFlow 2.x and TensorBoard
• Preparing the Input of Chatbots with Restricted Boltzmann Machines (RBMs) and Principal Component Analysis (PCA)
• Setting Up a Cognitive NLP UI/CUI Chatbot
• Improving the Emotional Intelligence Deficiencies of Chatbots
• Genetic Algorithms in Hybrid Neural Networks
• Neuromorphic Computing
• Quantum Computing
• Answers to the Questions
• Other Books You May Enjoy
• Index