These materials are © 2018 John Wiley & Sons, Inc.

Any dissemination, distribution, or unauthorized use is strictly prohibited.

 Deep
Learning
HPE Special Edition

by Scott D. Lowe

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Deep Learning For Dummies®, HPE Special Edition

Published by
John Wiley & Sons, Inc.
111 River St.
Hoboken, NJ 07030-5774
www.wiley.com
Copyright © 2018 by John Wiley & Sons, Inc., Hoboken, New Jersey

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any
form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise,
except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without
the prior written permission of the Publisher. Requests to the Publisher for permission should be
addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ
07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permissions.
Trademarks: Wiley, For Dummies, the Dummies Man logo, The Dummies Way, Dummies.com,
Making Everything Easier, and related trade dress are trademarks or registered trademarks of
John Wiley & Sons, Inc. and/or its affiliates in the United States and other countries, and may not be
used without written permission. All other trademarks are the property of their respective owners.
John Wiley & Sons, Inc., is not associated with any product or vendor mentioned in this book.

LIMIT OF LIABILITY/DISCLAIMER OF WARRANTY: THE PUBLISHER AND THE AUTHOR MAKE NO

REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE ACCURACY OR COMPLETENESS OF
THE CONTENTS OF THIS WORK AND SPECIFICALLY DISCLAIM ALL WARRANTIES, INCLUDING
WITHOUT LIMITATION WARRANTIES OF FITNESS FOR A PARTICULAR PURPOSE. NO WARRANTY
MAY BE CREATED OR EXTENDED BY SALES OR PROMOTIONAL MATERIALS.  THE ADVICE
AND STRATEGIES CONTAINED HEREIN MAY NOT BE SUITABLE FOR EVERY SITUATION. THIS
WORK IS SOLD WITH THE UNDERSTANDING THAT THE PUBLISHER IS NOT ENGAGED IN
RENDERING LEGAL, ACCOUNTING, OR OTHER PROFESSIONAL SERVICES.  IF PROFESSIONAL
ASSISTANCE IS REQUIRED, THE SERVICES OF A COMPETENT PROFESSIONAL PERSON SHOULD BE
SOUGHT. NEITHER THE PUBLISHER NOR THE AUTHOR SHALL BE LIABLE FOR DAMAGES ARISING
HEREFROM. THE FACT THAT AN ORGANIZATION OR WEBSITE IS REFERRED TO IN THIS WORK
AS A CITATION AND/OR A POTENTIAL SOURCE OF FURTHER INFORMATION DOES NOT MEAN
THAT THE AUTHOR OR THE PUBLISHER ENDORSES THE INFORMATION THE ORGANIZATION
OR WEBSITE MAY PROVIDE OR RECOMMENDATIONS IT MAY MAKE.  FURTHER, READERS
SHOULD BE AWARE THAT INTERNET WEBSITES LISTED IN THIS WORK MAY HAVE CHANGED OR
DISAPPEARED BETWEEN WHEN THIS WORK WAS WRITTEN AND WHEN IT IS READ.

ISBN 978-1-119-48458-5 (pbk); ISBN 978-1-119-48453-0 (ebk)

Manufactured in the United States of America

10 9 8 7 6 5 4 3 2 1

For general information on our other products and services, or how to create a custom For Dummies
book for your business or organization, please contact our Business Development Department in the
U.S. at 877-409-4177, contact [email protected], or visit www.wiley.com/go/custompub. For
information about licensing the For Dummies brand for products or services, contact BrandedRights&
[email protected].

Publisher’s Acknowledgments

Some of the people who helped bring this book to market include the following:

Development Editor: Business Development

Elizabeth Kuball Representative: Karen Hattan
Copy Editor: Elizabeth Kuball Production Editor:
Executive Editor: Katie Mohr Tamilmani Varadharaj

Editorial Manager: Rev Mengle

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Table of Contents
INTRODUCTION................................................................................................ 1
About This Book.................................................................................... 1
Foolish Assumptions............................................................................. 1
Icons Used in This Book........................................................................ 2
Beyond the Book................................................................................... 2

CHAPTER 1: Artificial Intelligence: All Buzz or the

Key to Your Reality...................................................................... 3
Defining Artificial Intelligence.............................................................. 4
Defining Machine Learning.................................................................. 5
Defining Deep Learning........................................................................ 8
Seeing What’s Driving Artificial Intelligence and
Deep Learning..................................................................................... 10
Making Artificial Intelligence and Deep Learning Practical
and Attainable..................................................................................... 11

CHAPTER 2: Considering IT Challenges with Deep

Learning Adoption..................................................................... 13
Suboptimal Infrastructure.................................................................. 13
Inability to Scale and Integrate.......................................................... 14
Limited Knowledge, Resources, and Capabilities............................ 14

CHAPTER 3: Demystifying Deep Learning with HPE..................... 15

Rapidly Developing Deep Learning Models..................................... 15
Scaling and Integrating Artificial Intelligence and
Deep Learning Capabilities................................................................ 17
Optimizing the Environment to Unlock Competitive
Advantage............................................................................................ 19

CHAPTER 4: Developing Deep Learning Models............................... 21

Selecting Ideal Technologies and Systems....................................... 21
Training Your Teams Using the Deep Learning Institute............... 24
Understanding Sample Artificial Intelligence Solutions................. 24

Table of Contents iii

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 CHAPTER 5: Delivering Attractive Returns and Results............ 27
Improving Deployment Effectiveness............................................... 27
Accelerating Analytics to Improve Business Operations................ 28
Integrating Artificial Intelligence and Deep Learning in
Your Environment............................................................................... 29
Seeing How Artificial Intelligence and Deep Learning
Impact Industries................................................................................ 30

CHAPTER 6: Enhancing Your Competitive Advantage................ 33

HPE Hardware Platforms.................................................................... 33
How Accelerated Graphics Processing Unit Compute
Impacts Learning Speed..................................................................... 39

CHAPTER 7: Ten (Or So) Ways HPE Can Help You

Get Started....................................................................................... 41

iv Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Introduction
S
elf-driving cars! Automation! Robots taking over jobs! These
are just a few of the things that the media covers when the
topic of artificial intelligence (AI) comes up. Of course, there
is far more to the story, and there are far more everyday ways that
the burgeoning fields of AI, machine learning, and deep learning
can transform your business, your career, and your life.

As the world becomes more intertwined with technology, ­digital

transformation is rapidly reshaping every industry. Some ­industries
have already transformed, while others  — like retail banking,
government, and manufacturing  — are under ­tremendous pres-
sure to innovate just to survive. AI plays a huge role in the era of
digital transformation.

Understanding the full scope of this ultra-hyped field will soon

be a make-or-break component for many careers and businesses.
Companies that fail to stay current with the AI wave will find
themselves left in the dust as competitors surge past them.

About This Book

AI and its components have the potential to reshape society as we
know it. In this book, you learn about all the areas that ­comprise
the AI field and gain an understanding for where each plays a
significant role. I also bust some myths about AI and provide you
with comprehensive information about how HPE’s products and
services can help companies like yours prepare for this major
change in the way that business is done.

Foolish Assumptions
In writing this book, I assumed that you have at least a basic
understanding of data center computing and virtualization. The
general audience for this book is anyone in IT who may want to
learn more about data center architectures. You may be part of
your organization’s technical staff, or you may be managerial or
executive staff. Either way, this book is for you!

Introduction 1

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Icons Used in This Book
Throughout this book, you find a number of icons intended to
help you better understand and remember key concepts.

The Remember icon points out anything that’s important enough

to commit to your long-term memory.

Anything marked with the Tip icon helps you save time or money
or just generally makes your life a little easier.

I don’t dive super deep into technical stuff in this book, but there
are some technical elements. I’ve marked them with the ­Technical
Stuff icon.

Beyond the Book

For more information on Deep Learning with HPE, visit www.
hpe.com/info/deep-learning.

2 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 IN THIS CHAPTER
»» Defining artificial intelligence, machine
learning, and deep learning

»» Understanding why artificial intelligence

and deep learning are popular

»» Adopting artificial intelligence

Chapter  1
Artificial Intelligence:
All Buzz or the Key to
Your Reality

S
o, what’s all the buzz around artificial intelligence (AI)? AI
has been showing potential for years, but interest seems to
have absolutely exploded recently. Analyst firm IDC predicts
that global spending on cognitive and AI solutions will increase at
a 54.4 percent compound annual growth rate (CAGR) over the
next few years, exceeding $46 billion by 2020. That’s a lot of
money.

Clearly, no business can — or should — avoid the AI reality. To

run a business effectively, you need information and intelli-
gence. One of the key reasons that AI has attracted a high level of
­corporate interest is because it can discover patterns in big data
(and much, much more).

And that interest is growing. In fact, you interact with AI on a daily

basis. Every time you ask Apple’s Siri to call home or you order
Amazon’s Alexa to tell you the current weather, you’re directly
interacting with AI-based systems, at a personal level. These
­personal AI systems are good for relatively simple, ­repeatable
tasks, and they’re becoming far more common.

CHAPTER 1 Artificial Intelligence: All Buzz or the Key to Your Reality 3

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 For you and your business, though, AI may not yet be a significant
reality. You may be eager to realize the benefits of AI, includ-
ing saving your company time and money. For example, you may
want to deploy enhancements to your customer service processes.
Perhaps an AI-driven chat-bot answering a customer query in
just seconds is cheaper than a human answering in minutes,
­saving your company money while also making the customer very
happy. If you use a service such as Uber, you’re buying a service
from a company that uses machine learning to create efficient
pickup and routing services. Pinterest uses machine learning to
combat spam and to improve monetization. These are things that
people use every day.

One of the biggest challenges people face when starting a ­journey

down the AI road is understanding what, exactly, is meant by
terms like artificial intelligence, machine learning, and deep learning.
The following sections help you make sense of the madness.

Defining Artificial Intelligence

AI is the capability of a machine to imitate intelligent human
behavior. AI is not a new concept. In fact, since the beginning
of the information age in the 1950s, we have experienced AI in
many forms. For example, early on, computer programs allowed
systems to play games like checkers against human ­opponents.
Today, Libratus, an AI, defeated four top human ­specialists in a
120,000-hand competition of heads-up no-limit Texas hold’em,
the leading benchmark and long-standing challenge problem in
imperfect-information game solving.

The term artificial intelligence is used to describe the broad field

and, itself, can have many different meanings and uses.

The original goal around AI was to create machines with the

characteristics of human intelligence. These thinking, ­reasoning
machines would be able to solve complex problems across the
spectrum of human capabilities, with access to replicas of all our
senses as inputs.

Today, we are realizing the promise of AI using machine and deep

learning techniques in our everyday lives, with examples like
Amazon’s Alexa, Google Maps, and Apple’s Siri.

4 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Defining Machine Learning
Simple mimicking of human behavior isn’t really all that hard.
Going beyond that requires a bit more effort, though, and that’s
where machine learning comes in. Machine learning is a subset
of artificial intelligence that gives computers the ability to learn
without being explicitly and rigidly programmed. The algorithms
that exist in a machine learning environment enable systems that
can go beyond traditional strict programming processes.

How does this work in action, though? You’ve probably used an

automated telephone system that takes your voice response so
that you can be connected to the correct party when you call a
big company. And if you’re like me, you get frustrated at just how
easily such systems can be confused! They’re rigidly programmed
to respond to a strict voice pattern.

But what if that automated system could learn based on its own
experiences? As each caller dials in, this automated ­attendant,
while fulfilling user requests, keeps track of where it made
­mistakes and everything that users did, and then self-adjusts in
subsequent interactions. Or perhaps it looks for common patterns
and identifies ways to streamline the process.

This is the realm of machine learning and there are a couple of

different ways that it can be accomplished.

Computers are taught, or self-learn, how to recognize things —

for example, the following:

»» Animals in pictures
»» The lanes on a freeway
»» A computer hacker’s digital fingerprints
»» Predict hard drive failures
»» Lung cancer from 3D image
There are innumerable uses for machine learning when the
­computer has figured out how to do the recognition, also referred
to as the training. This initial training can be very hard work, both
for the humans who prepare the training data and for the ­computer
that’s trying to create the models based on this ­training data.

CHAPTER 1 Artificial Intelligence: All Buzz or the Key to Your Reality 5

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 AI BRINGS LIFE TO COMPUTERS
When you step back and think about it, the machine learning process
is really a lot like our own development process, even if we don’t
remember it actually taking place! As the machine gains experience,
it’s able to more efficiently learn new capabilities, just like human
beings. The more life experience you have, the more of a baseline
you have from which to work.

But a ton of research is going into making the machine learning

training phase easier. For example, right now, it takes ­millions of
examples to train a machine learning system to identify what a
bird looks like. We humans can do it much, much more quickly, at
least for now. As you train the machine, it becomes more efficient
and, in fact, learns faster with iteration.

After the machine learning has completed its initial training, it

uses ongoing data to continually update its models — it’s always
learning. For example, a system that predicts the failure of water
pumps may be successful in those predictions 92 percent of
the time. But in 8 percent of cases, it either misses a failure or
­predicts a failure when none is imminent. These are considered
either false negatives or false positives. For these failures, the
system begins to ask itself some questions, including, “What is
it about my learning model that causes these false predictions,
these anomalies?” Hopefully the system can learn and improve.
Ideally, this learning can happen independently, but when it
can’t, people who understand machine learning can give it a
boost.

There are different methods by which machines can learn. I cover

these methods in the following sections.

Supervised learning
In supervised learning, machines are programmed with training
data sets, which ensure that the output is of a certain value or
quality. Supervised learning is, today, the most common way that
machines are trained.

6 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 If you didn’t think there were already enough challenges to
explore in the world of AI, supervised learning is itself broken
down into two major categories:

»» Classification: In classification-based supervised learning,

machines are taught to classify or categorize certain inputs
based on their programming. The outcome is often a state
or a characteristic. For example, if you’re teaching a machine
how to identify birds, classification could be “This is a bird” or
“This is not a bird.” It’s possible for more complex classifica-
tion processes to be implemented using more advanced
algorithms.
»» Regression: Regression learning allows a machine to
estimate values, such as future 401(k) earnings, based on
the input data you provide the machine. Rather than a
simple categorization, you get value-based results. The
more data you feed the machine, the more it can learn
and the more accurate it can be with its results.

Unsupervised learning
There’s a lot more to automated security than just monitoring for
known breach vectors, and that means that machine ­learning tools
sometimes need to act independently of their human ­supervisor.
Attackers are always adjusting their methods because they know
that simple systems can’t keep a watchful eye on everything all
the time.

Cyber-attacks are similar to the attacks on the human immune

system. All kinds of viruses attacking our bodies and morphing
all the time. And just like with the human immune system, the
defenders must also morph, too.

In simple terms, unsupervised learning starts with an algorithm

and a data set. Supervised learning also includes an output variable
of some kind, but unsupervised learning relies on the machine to
discover its own meaning in the data you provide it.

Not every tough problem is the same. There are a couple of

­categories in the world of unsupervised AI learning as well:

»» Clustering: In a clustering problem, you rely on the machine

to identify common characteristics in your underlying data.

CHAPTER 1 Artificial Intelligence: All Buzz or the Key to Your Reality 7

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 For example, you may cluster your client database based on
which services they’ve used over the past year. From a
business perspective, you may use this information to target
those customers for new services.
»» Association: Association-based machines look for ways to
discover data patterns that may not be that obvious. For
example, you may discover that most of the clients that
purchase one service also purchase a second service. With
this information, you may discover that bundling these
services increases your revenue as more customers buy
the bundle. That’s just a simple way that AI can help improve
your business.

Defining Deep Learning

Machine learning is certainly powerful, but deep learning takes
it to a whole new level. Deep learning is an evolution of machine
learning that creates increasingly complex hierarchical models
intended to mimic thought processes in the human brain more
than simple machine learning models.

As such, deep learning is a generation beyond machine learning,

while still being a subset of both AI and an evolution of machine
learning. Like so many technologies people have never heard of,
deep learning may already have an impact on your life.

People are delighted to discover that Netflix knows what they

want to watch. They gaze in amazement as Amazon displays other
products that they may want to buy.

Although magical, this isn’t the work of the black arts. These
kinds of recommendation engines are driven by deep learning
systems that can act on vast quantities of data to help improve
the quality of your life.

Machine learning is the mainstream today, but deep learning has

the most transformative potential to solve real-world problems
that go far beyond simple entertainment. Figure 1-1 sums up the
AI landscape.

8 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 FIGURE 1-1: A visual comparison of AI technologies.

To summarize, with rule-based AI and machine learning, a data

scientist has primary responsibility for deciding which rules and
data set features will be included in models, which drives how
those models operate. With deep learning, data scientists submit
raw data into an algorithm. The system analyzes that data and,
based on what it knows and what it can infer from the new data
you’ve presented, it makes a prediction.

ARTIFICIAL INTELLIGENCE AND

HIGH-PERFORMANCE COMPUTING
High-performance computing (HPC) clusters have been around for
ages, but they’ve found new customers thanks to the insatiable thirst
of AI. HPC and AI are like peanut butter and chocolate. Separate,
they’re both really tasty, but together, you get a special treat that will
delight your senses!

Traditionally, HPC systems have addressed complex data problems.

With the ability to support large data sets, in-memory databases, and
complex models, HPC solutions have the horsepower necessary to
help organizations gain insight into their data.

As the foundation for cognitive computing, organizations seeking

to harness AI to help meet their goals need advanced technology
­solutions, such as HPC. These solutions enable AI applications to
­execute increasingly difficult tasks and forecast evolving trends,
equipping them to solve some of the world’s biggest scientific,
­engineering, and technological problems.

CHAPTER 1 Artificial Intelligence: All Buzz or the Key to Your Reality 9

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Seeing What’s Driving Artificial
Intelligence and Deep Learning
No matter how incredible a technology appears to be, it won’t
gain any traction if there aren’t clear business outcomes that can
be attained. Deep learning is no different. Thanks to the potential
to surpass the competition, though, there are a number of reasons
that deep learning is going mainstream.

Gaining insight from increasing

­quantities of data
Data is the new currency. Whoever has the data will win the day.
Deep learning crunches data in ways that simple rules-based AI
systems can only dream. This insight can be directly respon-
sible for increased revenue generation or for improving cancer
­detection or for averting crop diseases in agriculture, and for so
much more.

Understanding the business

potential of deep learning
As a bank, how do you know — really know — that the customer
that you’re prepared to loan money to will pay it back? You have
his credit score to work with, which provides you with a good
snapshot of his credit profile, but what if you could do more? What
if you had a deep learning–backed AI system that could scour the
applicant’s history and discover that he’s a fraudster? You might
save your company a lot of money!

Perhaps even more significant is the reverse. Maybe an ­applicant

with a terrible credit score is actually a decent risk. A simple credit
score doesn’t tell the whole story, but with an automated ­system
sitting next to you, crunching data in an effort to provide you
with a more complete picture, you may decide that the appli-
cant is worth the risk and provide the loan. At the same time,
your ­system may discover that the applicant is suitable for other
products you make available. With the right data and the right
insights, you have the potential to dramatically enhance your
organization’s bottom line.

10 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 In agriculture, how do you leverage data to detect crop diseases so
that you can remove them before they create widespread damage?
Deep learning algorithms in agriculture are already doing exactly
this, resulting in increased crop yields.

In a manufacturing environment, deep learning is being put to

use to enable voice-based picking of warehouse parts. Voice is
a far more efficient and natural way to interact with inventory
­control systems, so these types of deep learning systems can
reduce costs and improve accuracy.

Making Artificial Intelligence and Deep

Learning Practical and Attainable
AI and deep learning are technologies that may seem like science
fiction to you. We all want a Commander Data by our sides, but
instead we get a rules engine to configure. The right mix of soft-
ware, hardware, and services is critical to leveraging the nascent
capabilities of deep learning and modern AI. The best part is that
all the ingredients for an incredible deep learning solution are
available to you today, and adoption may not be the pipe dream
you thought it would be.

See how HPE is helping customers navigate their AI journey

through the development of a three-phase process: Explore,
Experiment, and Expand.

Explore
If you haven’t discovered the real potential of using AI, now is the
time to jump in and explore. During this exploratory phase, you
need to determine what’s possible and how it may impact your
business. This is also where you can test the limits of the AI hype
cycle to see what’s real and what’s not. It’s also a stage at which
engaging a trusted partner such as HPE can pay huge dividends.
As an advisor, HPE can help you understand your options and,
with you, create an adoption plan that considers the infrastruc-
ture you need to get started.

CHAPTER 1 Artificial Intelligence: All Buzz or the Key to Your Reality 11

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Experiment
Now, play! What are the limits for what you’ve put into place?
How do you make it self-aware enough that it identifies and
assists with the critical aspects of your business and makes you
more competitive?

HPE can be your partner in your AI journey. Through the

­company’s Centers of Excellence, you can accelerate your deep
learning applications and realize better return on investment
(ROI) from your deep learning deployments in the near term.

Expand
After you’ve completed your experiments, try it in production,
where you can begin to realize the ROI that makes deep ­learning
such a compelling technology. As you discover more services
or opportunities that can benefit from AI, you can expand your
deep learning footprint to encompass these new use cases. Here,
too, HPE can help you implement and grow your environment to
­support even the most complex services you may dream up.

12 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 IN THIS CHAPTER
»» Seeing what’s holding you back from
realizing your machine-driven future

»» Getting help with implementing deep

learning

Chapter  2
Considering IT
Challenges with Deep
Learning Adoption

A
s you consider venturing down the deep learning road, you
probably anticipate running into road blocks along the
way. In this chapter, I fill you in on some of the key IT
challenges that can prevent companies from realizing the full
potential of implementing deep learning solutions.

Suboptimal Infrastructure
Unfortunately, traditional infrastructure doesn’t meet the
extreme compute requirements that are needed for deep learning
capabilities.

Unlike traditional applications, deep learning solutions require a

lot of parallelism, performance processors, dense storage, a lot of
power, and a lot of cooling. Traditional infrastructure may need
those things as well, but one of the main ways in which deep

CHAPTER 2 Considering IT Challenges with Deep Learning Adoption 13

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 learning infrastructure is different from traditional infrastructure
is that deep learning relies on high-performance accelerators
such as graphics processing units (GPUs).

Inability to Scale and Integrate

Let’s suppose you get past the infrastructure hurdle. After you’ve
developed a deep learning model and conducted your very first
proof of concept (POC), you want to get it into production as
quickly as possible. You want to get value out of your deep learn-
ing investment. You want to move from a POC to production, but
you may find yourself struggling with your ability to scale and
integrate.

You may lack the resources and expertise necessary to move out
of POC into a production environment. Many projects have these
kinds of challenges and require a boost to get beyond that, to a
point at which the business is benefitting.

An AI project that is stalled won’t yield any results. You may be

stalled going into production. You may be stalled in a learning
model. Your AI may be stalled at its current level of development
due to a lack of infrastructure resources. HPE is your proven part-
ner, here to get you out of being stalled and on your way to faster
insights and competitive advantage.

Limited Knowledge, Resources,

and Capabilities
As an emerging field, AI skills can be difficult to procure. To get
the right skills in a direct hire can be expensive and challenging.
To expand your AI footprint, you may need help optimizing your
environment, leveraging the latest in technologies, and you may
need to partner with a vendor to help. Fortunately, HPE has the
capability to help you leverage the latest technologies to gain com-
petitive advantage from your deep learning deployment. From the
company’s AI Innovation Centers to HPE Pointnext deployment
services, HPE can bring skills and experience to bear and, when
coupled with the right infrastructure, can help customers bring AI
to their lines of business.

14 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 IN THIS CHAPTER
»» Streamlining artificial intelligence
training models

»» Discovering resources that can help you

on your deep learning journey

Chapter  3
Demystifying Deep
Learning with HPE

D
eep learning is a complex topic. Many people simply don’t
know where to begin with the technology, let alone under-
stand what they can ultimately do with it. That’s where
HPE comes in. Through a combination of products, software, and
services, HPE can help companies decode deep learning to help
them benefit.

In this chapter, I walk you through how HPE can help.

Rapidly Developing Deep

Learning Models
Developing models can be challenging. Sometimes, you just need
help to get you past this initial level. That’s where HPE comes in.

Leveraging deep learning IT expertise

Deep learning workloads have unique requirements, particularly
when it comes to training modules. Choosing an ideal technology
platform isn’t simple.

CHAPTER 3 Demystifying Deep Learning with HPE 15

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 HPE can help IT departments create the optimal foundation for
deep learning with rich expertise and tools to select the right
combination of central processing unit (CPU) and performance
accelerators, like graphics processing unit (GPU) technologies,
and system platforms to meet their unique requirements. Many IT
departments, data scientists, and line of business decision mak-
ers lack a good understanding of deep learning frameworks and
infrastructure requirements.

A deep learning framework provides the building blocks for the

creation and training of deep learning systems. A number of such
frameworks are available, including the Microsoft Cognitive Tool-
kit, TensorFlow, Caffe, and many more. Each framework carries
with it a different set of operations. With the number of frame-
works available, choosing the right one for your needs can be
confusing. Some of the frameworks are easier to use and learn
than others as well. To learn more about some of the more com-
mon frameworks, go to http://hpe.to/6004DdXVK.

HPE can help to train data scientists and IT staff on the vari-
ous deep learning frameworks and the supporting infrastructure
requirements with hands-on or online training.

For those data scientists who are looking to gain quick access to
an ideal platform to develop deep learning models, HPE has sev-
eral options to enable rapidly developed deep learning models.

Through a mix of comprehensive hardware, software, services,

and other resources, HPE makes it possible for any organization
to harness the power of AI.

Providing solutions
HPE has created one such integrated solution through the combi-
nation of the HPE Apollo 6500 Gen 10 System and Bright Computing
software. This is an ideal platform to develop deep learning training
models. It comes with the maximum GPU capacity and the deep
learning frameworks that data scientists and application develop-
ers can use to get started. It also provides significant flexibility and
choice in terms of both hardware and software.

HPE has found that many IT departments are under pressure

from CIOs and CEOs to embark on AI initiatives.

16 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 If you’d like to take a bigger step and use existing software appli-
cations, HPE can help you build solutions that can help with,
among other things, fraud detection in e-commerce applications.
This is just one example of solutions that can be deployed.

Scaling and Integrating Artificial

Intelligence and Deep Learning
Capabilities
As you begin your deep learning expedition, you’ll probably find
that you’re able to create some simple models on your own. But
as you move these models out of development and into produc-
tion, you may find challenges in scaling the infrastructure and
integrating the massive data volumes to deliver real returns to
the business.

Providing proven blueprints

In your journey to overcome these challenges, HPE has created
an entire practice around helping organizations across the globe
implement deep learning solutions.

Reference architectures
If you’re interested in building the infrastructure for a specific use
case, such as image classification, HPE provides reference con-
figurations with hardware platforms and sizing guides that will
help you to successfully scale production deployments.

But hardware alone does not a functional AI cluster make. Man-

aging such clusters requires specialized tools. To this end, HPE
makes available cluster management software to scale manage-
ment for large deployments.

AI Innovation Centers
HPE makes available two key resources to help customers in their
AI journeys. The first, the HPE AI Innovation Center, is designed
for longer-term research projects. The Innovation Center serves as
a platform for research collaboration between universities, enter-
prises on the cutting edge of AI research, and HPE researchers.

CHAPTER 3 Demystifying Deep Learning with HPE 17

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 The centers, located in Houston, Palo Alto, and Grenoble, give
researchers for academia and enterprises access to infrastructure
and tools to continue research initiatives.

The second resource, HPE Centers of Excellence (CoE), are

designed to assist IT departments and data scientists who are
looking to accelerate their deep learning applications and realize
better ROI from their deep learning deployments in the near term.
The HPE CoE offers select customers access to the latest tech-
nology and expertise, including the latest NVIDIA GPUs on HPE
systems. The current CoE are spread across five locations includ-
ing Houston; Palo Alto; Tokyo; Bangalore, India; and Grenoble,
France.

Here are just some of the ways that the AI Innovations labs and
Centers of Excellence can help you get started with your deep
learning projects:

»» Driving technology development: Integrate technologies

required for deep learning and AI more quickly.
»» Demonstrating insightful benchmarking and testing:
Understand how solutions perform before fully investing.
»» Developing reference architectures and configurations:
Identify solutions specific to your requirements.
»» Validating and developing software: Prove that the
software performs as expected.

Both internal HPE researchers and leading researchers in acade-

mia working on AI projects need a stable, AI sandbox. The AI sand-
box is a testing environment that isolates untested code changes
and outright experimentation from the production environment
in the context of software development. They need access to sig-
nificant but not excessive hardware with comprehensive AI envi-
ronment while avoiding the need to re-provision for each use.
HPE has resources that can help.

Deploying, integrating, and supporting

a deep learning environment
Most organizations can’t just go it alone with AI. To assist, HPE
provides comprehensive services for deployment, data integration,

18 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 and support. Specifically, HPE Pointnext provides a wide array of
services around deep learning and AI, including the following:

»» Advisory services: With AI experts, HPE Pointnext can point

you in the right direction to help realize quicker time-to-value
from your AI investment.
»» Professional and operational services: Are you looking for
on-time implementation and on-budget execution for your
AI project? HPE Pointnext provides you with full-spectrum
support to ensure that both happen.

Bringing flexible, on-demand capacity

to the deep learning game
With the HPE GreenLake Flex Capacity service, HPE brings
increased financial flexibility and control to your on-
premises AI infrastructure deployment, which carries the follow-
ing characteristics:

»» HPE owns and supports your choice of on-premises

equipment.
»» Pay for what you use based on metered usage, and access
self-service reporting to forecast future demand.
»» An on-site buffer allows you to quickly scale up capacity
when needed and avoid spending on unused capacity.

Optimizing the Environment to Unlock

Competitive Advantage
After you successfully develop deep learning models and get them
into production, there is an opportunity to leverage the latest
technology innovations to accelerate performance and optimize
environments that can deliver significant competitive ­advantage.
HPE products and services are well positioned to make this
happen.

CHAPTER 3 Demystifying Deep Learning with HPE 19

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Surpassing your competition with
global centers of excellence
HPE’s Centers of Excellence help you leverage the latest tech-
nologies, modernize your application code, and demonstrate
the outcomes. In these centers, you can also perform proof-of-
concept activities and benchmarking. These centers of excellence
are customer facing and meant for larger customer engagements
where they can get access to expertise and tools.

Leveraging next-gen technology

integration
HPE doesn’t do all this alone. The company partners with top-
tier solution providers delivering core technologies that drive the
solution. HPE tests and verifies these platforms to ensure that
you’re able to gain the best in terms of next-gen technology
expertise.

20 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 IN THIS CHAPTER
»» Choosing the right technologies and
systems for you

»» Finding out how the Deep Learning

Institute can help

»» Looking at sample artificial intelligence

solutions

Chapter  4
Developing Deep
Learning Models

L
ots of resources are available to you as you begin developing
your own deep learning models. This chapter shows you
where to find the resources you need.

Selecting Ideal Technologies and Systems

The success of your deep learning initiative is dependent on the
robustness of the underlying technology. To help you get started
creating models right away, HPE has a number of resources you
can use.

Understanding HPE’s Deep

Learning Cookbook
The HPE Deep Learning Cookbook is a comprehensive set of tools
to guide the choice of the best hardware and software environment

CHAPTER 4 Developing Deep Learning Models 21

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 for different deep learning workloads. It consists of three main
components:

»» Deep Learning Benchmarking Suite: A benchmarking tool

to collect performance measurements on various hardware
and software configurations in a unified way
»» Deep Learning Performance Guide: A web-based tool to
analyze performance data collected with the Benchmarking
Suite, and to predict performance of untested hardware and
software configurations with analytical performance models
»» Reference Designs: Default hardware and software recipes
for select deep learning workloads, created based on
extensive benchmarking and performance analysis for these
workloads

Today’s benchmarking suite supports a wide range of the most

popular deep learning frameworks (BVLC Caffe, NVIDIA Caffe,
Intel Caffe, Caffe2, TensorFlow, MXNet, PyTorch and NVIDIA
inference engine TensorRT) and 18 popular deep models as refer-
ence workloads (see the full list at http://hpe.to/6008DdXzk).
Using the benchmarking suite, HPE has collected massive
amounts of performance data on multiple HPE hardware systems,
including HPE Apollo 6500, HPE Apollo 6000, HPE Apollo 2000,
HPE Edgeline 4000 and 1000, and HPE ProLiant DL380. This per-
formance data is publicly available through HPE’s Deep Learning
Performance Guide, a second component of the Cookbook.

So, what exactly is the Cookbook? It’s a combination of an auto-

mated benchmarking tool (to measure performance of different
hardware and software stacks), an analysis tool, and a set
of default hardware configurations for selected deep learning
workloads (Reference Designs). You use the benchmarking tool
to measure and validate the overall performance of your envi-
ronment. Then you use the web-based analysis tool to leverage
a massive collection of performance results for different deep
learning workloads on different hardware and software stacks
and analytical performance models.

There are key use cases that the Cookbook can help support — it’s
not just for newbies. For example, maybe you started at a small
scale and you’ve seen initial success. Now you want to scale up

22 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 the infrastructure. Scaling up is an altogether different problem.
HPE’s Deep Learning Cookbook helps you understand perfor-
mance bottlenecks and design a rock-solid scale-up, scale-out
architecture.

Or, maybe you started in the public cloud, but you want to bring
your AI deployment back on-premises. You may have chosen the
public cloud to get started and seen some initial success. Now
the line of business or the CIO wants to bring back the algorithms
on-premises to get the best cost, security, performance, and
­scalability. The Deep Learning Cookbook can help you design the
optimal architecture to bring back the model on-premises.

Getting expert advice to

jumpstart your initiative
Let’s dive deeper into the HPE Deep Learning Cookbook in action,
helping customers meet their goals.

Trying to get started, but don’t know how?

If you’re like most people, you’re fascinated by AI.  You want to

get started, but you may lack the expertise and resources to build
the model and the right infrastructure to run it. Data scientists
want to focus on building the algorithm, rather than building the
hardware/software infrastructure to run it. The HPE Deep Learn-
ing Cookbook recommends easy get-started configurations. It also
recommends which algorithms to use for which applications.

Using the HPE artificial

intelligence toolset
The Deep Learning Cookbook helps with answering such ques-
tions as “Which deep learning framework do I need?” and “Which
GPU should I choose, and how many?” But there is a lot more that
HPE makes available, too:

»» Deep Learning Benchmarking Suite

»» Deep Learning Performance Guide
»» Reference Designs (http://hpe.to/6009DdXRF)

CHAPTER 4 Developing Deep Learning Models 23

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Training Your Teams Using the
Deep Learning Institute
Keeping yourself and your team current with the fast-moving
nature of deep learning is a critical component of your project.
You need to start with the basics and become enmeshed in the
communities that have been developed to bring this technology
to where it is today.

HPE and NVIDIA are providing training for developers, data

­scientists, and researchers. The NVIDIA Deep Learning Institute
(DLI) gives customers access to online labs and hands-on work-
shops that will prepare their workforces to utilize the latest deep
learning techniques and software frameworks.

Understanding Sample Artificial

Intelligence Solutions
You must have figured out that you don’t need to do deep learn-
ing on your own. If you have a well-defined use case, it’s entirely
­possible that HPE has already developed a sample model and solu-
tion that you can get and start to use immediately. No AI training
time needed! Via reference architectures, you can get started with
your AI initiatives and avoid much of the trial and error that is
often associated with new endeavors.

Identifying fraudulent patterns

One of the most important deep learning services on the market
revolves around protecting people from fraud. Every day, thieves
steal people’s credit cards and their very identities in order to
get money. Financial fraud is a quickly growing problem and one
that most people will probably face at some point in their lives.
Think of it like this: How often does your bank contact you to
tell you that they’ve detected possible fraudulent activity on your
credit card?

For example, American Express recently notified me that one of

my cards had been used at a gas station in California. I live in
Missouri and wasn’t traveling that week. Through a deep under-
standing of my normal activity, American Express was able to flag

24 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 a number of potentially fraudulent transactions and block some
of them.

Many companies involved in finance and credit use such systems,

and you may want to use something similar. The HPE product
portfolio includes a fraud detection platform system, which can
identify fraudulent activity via queries in seconds, reduce false
positive rates, avoid costly losses, and generate a very quick
return on investment.

Quickly exploring anomalies

The previous example is just one among many that can be cited
as using deep learning to identify anomalous behavior. This kind
of out-of-pattern detection becomes possible as data sets become
larger and more populated and as deep learning algorithms are
made available, learning what normal behaviors look like.

Once trained, these AI systems can then place a flag on an account

or on a patient’s health records so that a human can take a closer
look and, if necessary, follow up with the cardholder or the patient.

Enhancing employee and business

­productivity with deep learning
Many people are concerned about the potential for AI and auto-
mation to replace people. Although this may be a valid concern,
it increases productivity and human resources, which can be
deployed elsewhere. As AI and deep learning systems become more
powerful, they’re able to replicate the results for certain human
tasks, and they can do so far more quickly and cost effectively.

In fact, we’ve already seen this kind of shift with simple AI systems.
When was the last time that you called a big company and got a
phone tree? Companies used to have operators who answered every
call. Today, decision systems have largely replaced these people.
The remaining employees are then free to focus on more critical
business activities and can leave the minutiae to the machines.

Data is king. By using the right data at the right time, organiza-
tions can dramatically improve operations, making operational
improvements that have measurable, impactful results. With the
ability to analyze billions of data points in real time, deep learning
solutions can have an immediate benefit on how your business
does business.

CHAPTER 4 Developing Deep Learning Models 25

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 For example, can a chat-bot do a better job at handling initial
or simple customer queries? Can an AI-backed robot do a more
accurate job when it comes to fulfilling customer orders? Can a
deep learning–centric cyber defense system improve the overall
availability of your network infrastructure?

All of these examples and more are possible today. More impor-
tantly, these advances can help you delight your customers and
increase your profits.

26 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 IN THIS CHAPTER
»» Deploying the best possible deep
learning system

»» Streamlining your data-centric journey

»» Integrating AI into your environment

»» Looking at some use cases

Chapter  5
Delivering Attractive
Returns and Results

G
etting your deep learning projects underway is easier when
you have the right tools at your fingertips. In this chapter,
I let you know about even more tools that HPE offers to
help you successfully deploy AI in your business.

Improving Deployment Effectiveness

Building your deep learning environment requires significant
knowledge and experience of the needs of this emerging field. Fig-
uring out all of the pieces you need to make it work is challenging.

HPE has complete reference architectures that you can either use
directly or use as a baseline for your own efforts, and build a deep
learning environment that will solve your company’s challenges.

HPE’s deployment guides help you hit the ground running with
constructs that enable a new research-driven architecture to take
hold in your organization. With this architecture, you can start
to use deep learning to, for example, identify suspects in crimes,

CHAPTER 5 Delivering Attractive Returns and Results 27

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 perform tissue classification in digital pathology, ID vehicle
license plates, and much more.

HPE offers flexible deployment options for you to get started

right away. Choose from an on-premises but cloud-like deploy-
ment model where you pay as you go, or buy the infrastructure
outright  — whichever option makes the most sense for your
business.

Accelerating Analytics to Improve

Business Operations
The marriage of high-performance computing (HPC) and AI
has the potential to create offspring that are greater than the sum
of their parts. As these technologies continue to converge, orga-
nizations are seeking comprehensive solutions to make AI easier
to deploy and support.

One area that is of great interest is analytics. With the ability to

delve more deeply into data, organizations can realize massive
benefits. AI is helping financial services companies, but much
more is possible.

There are myriad opportunities for analytics to improve business

operations. Let’s look at an industry in which the data patterns
change, literally, every second: the weather. At its core, predicting
the weather is a big data and analytics problem. By bringing more
horsepower to bear, intelligence systems can more accurately
predict weather patterns, which can save both lives and money.
The earlier and more accurate weather predictions are, the easier
it is for you to minimize its impact on your life.

And that’s just for the general population. Now, imagine how that
same improvement in prediction impacts air travel and delivery
companies. With an improvement in weather prediction, these
organizations can more effectively and with less impact alter their
schedules and accommodate passengers.

Bringing HPC solutions with graphics processing unit (GPU)

underpinnings to the AI party makes it possible for improved
analytics to have a serious impact on business.

28 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Integrating Artificial Intelligence and
Deep Learning in Your Environment
Where do you start? You start with the business problem. If you
don’t know what problems you’re trying to solve, you certainly
won’t be able to solve them! Remember: Your deep learning out-
comes can be simple, or they can be complex.

Discovering powerful research

capabilities
With your business problems defined, it’s time to start dis-
covering how you can harness AI to solve them. Here, you
can start to “play.” You don’t necessarily need to build out a
complete environment. There are plenty of very simple deep learn-
ing applications you can start to work with using just your desktop
computer. Here’s an excellent resource that can point you in the
right direction so you can learn some of the basics behind discov­
ering some real-world potential: www.machinelearningmastery.
com/inspirational-applications-deep-learning.

When my kids were little, the motto was “learning through play.”
The same goes for everything you want to learn, AI and deep
learning included. So, start playing!

Deploying a stable artificial

intelligence environment
Ultimately, of course, after you’ve finished experimenting, you’ll
need to consider a more production-friendly environment than
your laptop. That said, you don’t need a lot to get started deploying
a stable AI environment. HPE has a complete line of deep
learning–friendly hardware resources, from the HPE ProLiant
DL380 Gen10 Server up through the HPE SGI 8600.

Streamlining deployment
No matter which HPE platform you choose, work with HPE expert
resources to make sure it’s suitable for your needs. HPE’s advi-
sory services will help you choose the right platform and software
to make even the most complex models work.

CHAPTER 5 Delivering Attractive Returns and Results 29

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Seeing How Artificial Intelligence and
Deep Learning Impact Industries
If you still aren’t convinced that AI and deep learning are impact-
ing entire industries, read on. But don’t be discouraged if your
needs are smaller in scope in comparison. Everyone has to start
somewhere!

Improving healthcare outcomes

The benefits of AI in the healthcare industry are huge. Nurses will
gain up to 50 percent more productivity using AI, patients will
realize up to 9 percent cost savings through custom treatments,
and most significant, the industry will see a $10 trillion savings
through advances in drug discovery (that’s trillion with a t). These
predictions are based on a McKinsey study for North America
healthcare cognitive computing market by technology that spans
2012 to 2022.

The overall potential outcomes with AI-guided healthcare are

many and are significant:

»» Increased accuracy of diagnoses

»» Improved preventive medicine and public health
»» Enhanced personalized care
»» Accurate prediction of insurance costs
»» Streamlined research and development with prediction
models
»» Guided drug development to deliver medications that meet
public need
»» Better patient outcomes overall

Accelerating cancer research

Although “cancer” as a monolithic disorder is unlikely to be
defeated in the near future, various types of cancer are being bat-
tled every single day inside the silicon wafers that power AI clus-
ters. Through powerful machine-based algorithms, researchers
are constantly discovering new vectors by which to fight various

30 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 cancers. On the patient front, AI-based systems are delving deep
into human bodies to help detect anomalies earlier in an effort to
ensure that treatment can begin before the disease has progressed
to an incurable stage.

Classifying tissue for cancer detection and

disease prevention
Earlier in this book, I talked to you about simple machine
­learning–based AIs that used robust data sets to, for example,
determine whether a photo that was being viewed contained a
picture of a bird. This simple yes-or-no algorithm compares
what it’s looking at with its vast repository of learned images and
makes a guess based on its prior knowledge.

Now, the same kind of power could be brought to bear in review-

ing human tissue and classifying it as either “healthy” or “not
healthy.”

Gaining insight in manufacturing

What does AI mean for manufacturing? In short, as today’s
manufacturers continue to be pressed to reduce costs, they’ll make
further investments in automation. At the same time, they need
to ensure that their lines keep running. A line that isn’t running
isn’t making money.

Manufacturers can turn to an AI-assisted tool that allows them

far greater insight into their operations than ever before. These
systems are being brought to bear on complex problems, such
as creating predictive maintenance schedules. With predic-
tive maintenance, manufacturing organizations can avoid costly
unplanned downtime thanks to an AI that lets management know
when something is about to go wrong.

Staying safe online

Between the bird photos and political fights on the Internet lie
some disturbing facts. One-third of all Internet traffic is consid-
ered malicious traffic intended to either defraud, disable, or dam-
age a victim. New malware is being released every few minutes.
And ransomware attacks continue to be massive moneymakers
for criminals. That’s not good.

CHAPTER 5 Delivering Attractive Returns and Results 31

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 The speed at which attacks take place and new malware is released
makes it impossible for human-driven cyber defenses alone to
take action. By the time a human has discovered a breach, attackers
may have made off with millions of sensitive data elements. If
not today, then in the not-too-distant future, AI systems will be
present in most organizations, watching potentially billions of
data points in an effort to get in front of malicious activity.

Speed is critical in cyber defense, but not at the cost of accuracy.

False positives from automated alerting systems can erode faith
in systems and lead human security pros down the wrong path.
Cyber defense systems will require very high levels of training
to ensure that the human side of the equation doesn’t end up
breaking down.

32 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 IN THIS CHAPTER
»» Discovering the hardware that makes
the magic happen

»» Finding out why GPUs are the

powerhouses of the AI world

»» Generating returns for buyers with

real-world AI-driven products

Chapter  6
Enhancing Your
Competitive Advantage

I
n the previous chapters, I talk about what artificial intelligence
(AI) and deep learning can do for you and provide generalities
around how HPE can help. In this chapter, I jump into some
specific products from HPE that will get you from zero to AI
­goodness in a nanosecond.

HPE Hardware Platforms

You know that you need hardware in order to operate your deep
learning environment. HPE provides a broad portfolio that takes
you from getting started through advancing your mature plat-
forms to allow them to further propel your business.

HPE Edgeline EL4000

The HPE Edgeline EL4000 (see Figure 6-1) is designed to ­perform
inference at the edge, regardless of environmental conditions.
This purpose-built solution can acquire data, analyze it on an
industry-standard x86 deep compute platform, drive high-
performance displays for operators, and initiate control actions
in real time — at the edge, removing transfer issues to result in
faster insights and business agility.

CHAPTER 6 Enhancing Your Competitive Advantage 33

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 FIGURE 6-1: The HPE Edgeline EL4000.

HPE ProLiant DL380 Gen10

HPE’s all-time best-selling server is the ProLiant DL380, cur-
rently in its tenth generation (see Figure 6-2). Although it’s usable
primarily for small-scale AI deployments, the DL380 can support
NVIDIA GPUs for AI training, and you can add multiple servers to a
cluster as you need to expand your machine learning capabilities.

FIGURE 6-2: The HPE ProLiant DL380 Gen10.

HPE Apollo sx40

The Apollo sx40 provides an extremely dense and powerful 1U GPU
server for deep learning. The sx40 can support up to four NVIDIA
GPUs, including the latest V100 GPUs with NVIDIA NVLink, allow-
ing it to provide vastly accelerated computing in a very dense
standard form factor, air-cooled package. The sx40 is ideal as
an entry-level GPU server to investigate, learn, and develop deep
learning skills and experience.

HPE Apollo 2000 Gen10

You know that scale is important in deep learning, but some-
times, you’re okay starting a bit smaller. The HPE Apollo 2000
Gen10 system (see Figure 6-3) is a perfect fit for those times. The
system supports up to four nodes per 2U chassis.

34 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 FIGURE 6-3: The HPE Apollo 2000 Gen10 Chassis.

Offering a flexible configuration, the HPE Apollo 2000 Gen10 sys-

tem supports a variety of workloads, from remote site systems to
large AI-centric HPC clusters to everything in between. You can
deploy the HPE Apollo 2000 Gen10 cost-effectively, starting small
with a single 2U shared infrastructure and scaling out up to 80
HPE ProLiant Gen10 servers in a 42U rack.

The HPE Apollo 2000 is quite versatile and can be used for train-
ing models in the data center and serve as an inference engine
on the edge. The ProLiant XL190r Gen10 Server node that goes
into the Apollo 2000 chassis is a 2U half-width, dual-processor
server with additional PCIe slots in multiple configurations pro-
viding support for additional expansion cards and support for two
accelerators per server.

HPE Apollo 6500 Gen10

The HPE Apollo 6500 Gen10 (shown in Figure  6-4) is a reliable
enterprise deep learning and HPC platform with industry-leading
accelerators delivering exceptional performance for faster intel-
ligence. With eight high-performance graphical processing units
(GPUs) per server, the system provides superior performance
per dollar for GPU-intensive workloads  — delivering up to 125
TFLOPs single-precision compute. Purpose-built for accelerated
computing, this platform features both the traditional PCIe and
the new NVIDIA NVLink GPU interconnects, providing the flex-
ibility to suit a wide variety of deep learning and HPC require-
ments. The NVIDIA NVLink GPU interconnect is particularly
useful for deep learning workloads, characterized by heavy GPU-
to-GPU communications.

CHAPTER 6 Enhancing Your Competitive Advantage 35

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 FIGURE 6-4: The HPE Apollo 6500 Gen10.

HPE SGI 8600

HPC and AI go hand in hand with one another. The HPE SGI
8600 is a clustered, high-performance computing (HPC) system
intended to provide your AI endeavor with the horsepower neces-
sary to meet even the most demanding needs.

For petaflop-scale deep learning, HPE introduced the HPE SGI

8600 (see Figure 6-5). This liquid-cooled platform delivers lead-
ing performance, density, and efficiency and is used for extremely
large training exercises. There are three compute trays, with the
HPE XA780i Gen10 Server as the ideal deep learning platform
used to monitor and observe data processing for pattern recog-
nition and anomaly detection. This tray features a two–central
processing unit (CPU) socket node with the Intel Xeon Scalable
processors with support for up to four NVIDIA Tesla graphics pro-
cessing units (GPUs) with NVLink.

FIGURE 6-5: The HPE SGI 8600.

36 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 A petaflop is a unit of system performance that equates to one
quadrillion operations per second. AI systems need that kind of
performance for real-time applications.

HPE AI storage solutions

Data is at the heart of deep learning. HPE provides different types
of data storage solutions to meet the capacity and performance
requirements of the various AI workloads, from pure capacity
storage to pre-processing applications to training models in a
GPU cluster. A hyperconverged approach offers fastest throughput
with direct access to GPUs, while detached storage allows for
flexibility to scale GPU and storage independently. HPE provides
flexibility and choice of storage options to fit your infrastructure
requirements.

WekaIO Matrix
WekaIO Matrix is a flash-optimized parallel file system for
large training sets and large GPU clusters. The WekaIO solution
provides high performance and low latency for faster data set
processing. With support for multiple protocols, such as POSIX,
S3, and HDFS, automatic data tiering, and cloud bursting capa-
bilities, WekaIO Matrix can flexibly support the entire deep learn-
ing dataflow and scale for performance and capacity growth as
system requirements evolve.

DDN
A hybrid scale-out parallel file system solution, DDN high-
performance storage solutions scale from test-bed to produc-
tion with a single platform. The DDN storage and flash solutions
can support the entire deep learning dataflow, including provid-
ing capacity, ingest capability, and flash burst buffer technology
for performance and responsiveness. By providing an integrated
hybrid environment and using flash and hard drives, DDN pro-
vides the ability to define a cost-effective scaling strategy based
on current requirements for either capacity or performance.

CHAPTER 6 Enhancing Your Competitive Advantage 37

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 HPE Apollo 4200 with Qumulo QF2
The HPE Apollo 4200 system with Qumulo QF2 offers an end-
to-end scale-out NAS solution for unstructured data, offering
capacity storage as well as linear performance scalability. Data
reliability is provided with block-level erasure coding and built-in
analytics, allowing administrators to pinpoint problems and
control how storage is used. Cloud capabilities with continuous
replication and snapshots offer additional scalability with cloud
economics.

HPE Scalable Object Storage

with Scality RING
A dense storage system for actively archiving large data sets, HPE
Scalable Object Storage with Scality RING is a more economic
storage tier as data gets colder.

HPE Scalable Object Storage with Scality RING provides the ideal
large-scale active archive storage tier for deep learning data. Deep
learning often requires large amounts of data, which then must
be labeled, processed, and iterated upon, which further increases
the amount of storage that is required post-processing. HPE Scal-
able Object Storage is an economic, dense storage solution that
provides reasonable access speed while serving as the base tier
archive for all deep learning applications.

HPE Data Management Framework

Deep learning data is not static. It gets cleaned, labeled, pro-
cessed, transformed, and iterated upon. This makes data man-
agement essential, because data files must be not only stored,
but also collected and moved to serve compute requirements. The
HPE Data Management Framework, working in concert with AI
storage platforms, simplifies data workflows, using an automated
policy engine to move data to the right place at the right time
for processing requirements, and then moves the results of pro-
cessing to the appropriate storage infrastructure to reduce overall
cycle times.

38 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 How Accelerated Graphics Processing
Unit Compute Impacts Learning Speed
Why use expensive GPUs when you could just use a bunch of
less expensive CPUs to operate your deep learning environment?
It comes down to how each is architected and operates.

There are a number of key architectural differences between

AI and typical enterprise applications. Enterprise applications
work well with standard CPUs because those applications aren’t
generally all that parallel in nature. AI, on the other hand, can be
a massively parallel operation. As a result, you could easily flood
an entire rack of CPU-based servers with an AI problem and still
not get the results you’d see from a good GPU.

Why? It’s all about the cores. CPUs might have, say, 20 cores.
GPUs, on the other hand, can have thousands. And because each
individual workload can be pretty small, you don’t need massive
cores to complete individual tasks. GPUs are purpose-built to
handle complex floating-point mathematics equations with great
precision.

CHAPTER 6 Enhancing Your Competitive Advantage 39

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 40 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 Chapter  7
Ten (Or So) Ways HPE
Can Help You Get
Started

I
t’s clear that getting started with artificial intelligence, deep
learning, and machine learning should be on your short list of
learning goals. Here are ten (or so) ways that HPE can help you
get underway:

»» Turn to HPE for help in making deep learning work for

you. HPE has several tools to help you overcome your IT
challenges and leverage the latest deep learning technolo-
gies to gain competitive advantage.
With the right mix of hardware, software, services, and
training, any organization can harness the power of AI.
Check out the following for more information:

• HPE Demystifies Deep Learning For Faster Intelligence:

www.nextplatform.com/2017/10/25/hpe-demystifies-
deep-learning-faster-intelligence/
• HPE Apollo 6500 for Deep Learning: www.youtube.
com/watch?v=wkI4CFKWxFs

CHAPTER 7 Ten (Or So) Ways HPE Can Help You Get Started 41

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 • Transformation workshop: https://h20195.www2.hpe.
com/v2/Getdocument.aspx?docname=a00042611enw
• Transformation services: www.hpe.com/us/en/
services/consulting.html#servicesPortfolio
• HPE Pointnext services: www.hpe.com/services
»» See how HPE can help you figure out your optimal recipe
for configuring your hardware and software. You can use
tools like the Deep Learning Cookbook, which includes a
benchmarking suite, the performance analysis tool, and
reference designs. You can also further educate yourself by
checking out the Deep Learning Workshops. Go to http://
dlpg.labs.hpe.com.
»» Discover HPE’s sample AI solutions from a variety of
verticals, which can help ease you into the world of deep
learning and assist you in getting results from your
investment into the technology.
»» Learn from the comprehensive resources that HPE has
made available for you, all intended to help you under-
stand how deep learning can be used in your organization:

• Deep Learning Workshops: https://h10076.www1.

hpe.com/ww/en/training/solution/deep_learning_
AI.html
• Deep Learning Cookbook: https://developer.hpe.
com/platform/hpe-deep-learning-cookbook/home
»» Find out how HPE is establishing itself as a leader in deep
learning technology and giving back to the community in
an effort to accelerate deployment and standardization.

• DL benchmarking suite: https://community.hpe.

com/t5/Behind-the-scenes-Labs/The-Deep-
Learning-Benchmarking-Suite-now-available/
ba-p/6989905#.WoIF2SXwaUk
• Comparative analysis of different DL solutions:
http://on-demand.gputechconf.com/gtc-eu/2017/
presentation/23245-sorin-cheran-hpes-deep-
learning-cookbook.pdf

42 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 »» Let HPE help you with your research via two key
resources:

• HPE AI Innovation Center: Schedule time at the HPE AI

Innovation Center, which is designed for longer-term
research projects. The Innovation Center serves as a
platform for research collaboration between universities,
enterprises on the cutting edge of AI research, and HPE
researchers. The centers — located in Houston, Texas;
Palo Alto, California; and Grenoble, France — give
researchers for academia and enterprises access to
infrastructure and tools to continue research initiatives

• Enhanced HPE Centers of Excellence (CoEs): Visit an

Enhanced HPE CoE for help from the best minds in AI. CoEs
are designed to assist IT departments and data scientists
who are looking to accelerate their deep learning applica-
tions and realize better return on investment (ROI) from
their deep learning deployments in the near term. The HPE
CoEs offers select customers access to the latest technology
and expertise, including the latest NVIDIA GPUs on HPE
systems. The current CoEs are spread across five locations:
Houston, Texas; Palo Alto, California; Tokyo, Japan;
Bangalore, India; and Grenoble, France.
»» See how HPE can help your AI project on the financial
front. Can’t quite afford the upfront spend you might
associate with a deep learning project? Discover HPE
GreenLake, HPE’s on-premises pay-as-you-go consumption
model. Pay only for what you use. Whether it’s infrastructure,
real-time analytics, big data, backup, or protecting apps and
data — business leaders now widely prefer to consume the
outcomes they want. Visit HPE GreenLake Flex Capacity at
www.hpe.com/services/flexiblecapacity.
»» Discover more help from HPE in the form of the HPE
Data Management Framework. The HPE Data Management
Framework, working in concert with AI storage platforms,
simplifies data workflows, using an automated policy engine
to move data to the right place at the right time for processing
requirements, and then moves the results of processing to
the appropriate storage infrastructure to reduce overall
cycle times.

CHAPTER 7 Ten (Or So) Ways HPE Can Help You Get Started 43

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
 »» Find out how HPE is going beyond its own walls to help
customers deploy the most comprehensive solution.
Discover HPE’s broad partner ecosystem for deep learning.
From fraud detection solutions to custom solutions built by
HPE Pointnext and HPE partners, there’s a solution for all of
your deep learning needs.

You’re now ready for your own deep learning journey to success.
You can

»» Contact an HPE sales rep to learn how you can start on your
journey today.
»» Download articles to augment your existing knowledge.
»» Visit the HPE Deep Learning website (www.hpe.com/info/
deep-learning).

44 Deep Learning For Dummies, HPE Special Edition

These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
These materials are © 2018 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
WILEY END USER LICENSE AGREEMENT
Go to www.wiley.com/go/eula to access Wiley’s ebook EULA.