AI for data science

Contents

1. Introduction

1. About AI

2. AI facilitates data science

3. About the book

2. Chapter 1: Deep Learning Frameworks

1. About deep learning systems

2. How deep learning systems work

3. Main deep learning frameworks

4. Main deep learning programming languages

5. How to leverage deep learning frameworks

6. Deep learning methodologies and applications

7. Assessing a deep learning framework

8. Summary

3. Chapter 2: AI Methodologies Beyond Deep

Learning

1. Optimization

2. Fuzzy inference systems

3. Artificial creativity

4. Additional AI methodologies

5. Glimpse into the future

6. About the methods

7. Summary

4. Chapter 3: Building a DL Network Using MXNet

1. Core components

2. MXNet in action

3. MXNet tips

4. Summary

5. Chapter 4: Building a DL Network Using

TensorFlow

1. TensorFlow architecture

2. Core components

3. TensorFlow in action

4. Visualization in TensorFlow: TensorBoard

5. High level APIs in TensorFlow: Estimators

6. Summary

6. Chapter 5: Building a DL Network Using Keras

1. Core components

2. Keras in action

3. Model Summary and Visualization

4. Converting Keras models to TensorFlow

Estimators

5. Summary

7. Chapter 6: Building an Optimizer Based on the

Particle Swarm Optimization Algorithm

1. PSO algorithm

2. Main PSO variants

3. PSO versus other optimization methods

4. PSO implementation in Julia

5. PSO in action

6. PSO tips

7. Summary

8. Chapter 7: Building an Optimizer Based on

Genetic Algorithms

1. Standard Genetic Algorithm

2. Implementation of GAs in Julia

3. GAs in action

4. Main variants of GAs

5. GA framework tips

6. Summary

9. Chapter 8: Building an Optimizer Based on

Simulated Annealing

1. Pseudo-code of the Standard Simulated

Annealing Algorithm

2. Implementation of Simulated Annealing in

Julia

3. Simulated Annealing in action

4. Main Variants of Simulated Annealing

5. Simulated Annealing Optimizer tips

6. Summary

10. Chapter 9: Building an Advanced Deep Learning

System

1. Convolutional Neural Networks (CNNs)

2. Recurrent Neural Networks

3. Summary

11. Chapter 10: Building an Optimization Ensemble

1. The role of parallelization in optimization

ensembles

2. Framework of a basic optimization ensemble

3. Case study with PSO Systems in an ensemble

4. Case study with PSO and Firefly ensemble

5. How optimization ensembles fit into the data

science pipeline

6. Ensemble tips

7. Summary

12. Chapter 11: Alternative AI Frameworks in Data

Science

1. Extreme Learning Machines (ELMs)

2. Capsule Networks (CapsNets)

3. Fuzzy logic and fuzzy inference systems

4. Summary

13. Chapter 12: Next Steps

1. Big data

2. Specializations in data science

3. Publicly available datasets

4. Summary

14. Closing Thoughts

15. Glossary

16. Transfer Learning

1. When is transfer learning useful?

2. When to use transfer learning

3. How to apply transfer learning

4. Applications of transfer learning

17. Reinforcement Learning

1. Key terms

2. Reward hypothesis

3. Types of tasks

4. Reinforcement learning frameworks

18. Autoencoder Systems

1. Components

2. Extensions of conventional autoencoder

models

3. Use cases and applications

19. Generative Adversarial Networks

1. Components

2. Training process

3. Pain points of a GAN model

20. The Business Aspect of AI in Data Science

Projects

1. Description of relevant technologies

2. AI resources

3. Industries and applications benefiting the

most from AI

4. Data science education for AI-related projects

21. Using Docker Image of the Book’s Code and

Data

1. Downloading the Docker software

2. Using Docker with an image file

3. Docker tips

22. Index

Landmarks

1. Cover

2 Lindsley Road

Basking Ridge, NJ 07920 USA https://www.TechnicsPub.com

Cover design by Lorena Molinari Edited by Sadie Hoberman and Lauren

McCafferty All rights reserved. No part of this book may be reproduced or

transmitted in any form or by any means, electronic or mechanical,

including photocopying, recording or by any information storage and

retrieval system, without written permission from the publisher, except for

the inclusion of brief quotations in a review.

The author and publisher have taken care in the preparation of this book,

but make no expressed or implied warranty of any kind and assume no

responsibility for errors or omissions. No liability is assumed for

incidental or consequential damages in connection with or arising out of

the use of the information or programs contained herein.

All trade and product names are trademarks, registered trademarks, or

service marks of their respective companies and are the property of their

respective holders and should be treated as such.

First Edition

First Printing 2018

Copyright © 2018 Yunus Emrah Bulut and Zacharias Voulgaris, PhD

ISBN, print ed. 9781634624091

ISBN, PDF ed. 9781634624121

Library of Congress Control Number: 2018951809

Introduction

There is no doubt that Artificial Intelligence (commonly abbreviated AI) is

making waves these days, perhaps more than the world anticipated as

recently as the mid-2010s. Back then, AI was an esoteric topic that was

too math-heavy to attract the average computer scientist, but now, it seems

to be a household term. While it was once considered sci-fi lingo, it’s now

common to see and hear the term “AI” featured in ads about consumer

products like smart phones.

This is to be expected, though; once an idea or technology reaches critical

mass, it naturally becomes more acceptable to a wider audience, even if

just on the application level. This level refers to what AI can do for us, by

facilitating certain processes, or automating others. However, all this often

gives rise to a series of misunderstandings. As AI itself has become more

well-known, so have spread various ominous predictions about potential

dangers of AI — predictions that are fueled by fear and fantasy, rather

than fact.

Just like every other new technology, AI demands to be discussed with a

sense of responsibility and ethical stature. An AI practitioner, especially

one geared more towards the practical aspects of the field, understands the

technology and its limitations, as well as the possible issues it has, which

is why he talks about it without hyperbole and with projections of

measured scope – that is, he projects realistic applications of AI, without

talking about scenarios that resemble sci-fi films. After all, the main issues

stemming for the misuse of a technology like this have more to do with the

people using it, rather than the technology itself. If an AI system is

programmed well, its risks are mitigated, and its outcomes are predictably

positive.

About AI But what exactly is AI? For

starters, it’s nothing like what sci-fi books

and films make it out to be. Modern AI

technology helps to facilitate various

processes in a more automatic and

autonomous way, with little to no

supervision from a human user. AI

initiatives are realistic and purely

functional. Although we can dream about

what AI may evolve into someday, as AI

practitioners, we focus on what we know

and what we are certain about, rather than

what could exist in a few decades.

AI comprises a set of algorithms that make use of information – mainly in

the form of data – to make decisions and carry out tasks, much like a

human would. Of course, the emulation of human intelligence is not an

easy task; as such, the AIs of today are rudimentary and specialized.

Despite their shortcomings, though, these modern systems can be

particularly good at the tasks they undertake, even better than humans. For

example, an AI system, which is a standalone program implementing one

or more AI algorithms, that is created for identifying words from speech,

can be more accurate than humans doing the same task.

It’s important to note that all the AI systems we have today possess what

is termed narrow artificial intelligence. This means that current AIs can

do a limited set of tasks (or even just a single task) quite well, but offer at

best mediocre performance at any other task. For instance, an AI might be

great at figuring out your age based on a headshot, but that same AI

almost certainly couldn’t tell a classical music piece from a pop song.

Some AIs are designed to be used in robots, such as those designed for

rescue missions, able to navigate various terrains. Other AIs are

specialized in crunching data and facilitating various data analytics tasks.