Course

Executive Briefing

FIVE EXCITING EMERGING FIELDS

Welcome to the course!
As you’ll discover over these five modules, This executive briefing course is made up of five
there is a wealth of new technologies that short courses, where you will:
have a varietyof business applications. These Get a bird’s eye view of the field
moduleswill teach you all about exciting Understand how each technology works
new fields thatare saving companies Learn why it’s one of the fields which can put your
millions of dollars andchanging lives. company lightyears ahead of the competition
See real-world applications of the technologies
This course covers the five most exciting,
emerging fields:
Computer Vision (CV)
Deep Learning (DL) And all of that in under 30 per module
Reinforcement Learning (RL) minutes! Excited? Then let’s jump in!
Natural Language Processing (NLP)
Robotic Process Automation (RPA)

i
Table of Contents

PAGE

01 Course 1
Computer Vision

What is computer vision? (pg. 2)

How does computer vision work? (pg. 3)
Applications of computer vision (pg. 4)
Real-world use cases (pg. 5)

06 Course 2
Deep Learning

What is Deep Learning? (pg. 7)

Building Neural Networks (pg. 8)
Deep Learning and other areas of AI (pg. 9)
Real-world use cases (pg. 10)

11 Course 3
Reinforcement Learning

What is Reinforcement Learning? (pg. 12)

Reinforcement Learning in Humans (pg. 13)
Advantages of Reinforcement Learning (pg. 14)
Reinforcement Learning in Marketing (pg. 15)

ii
16 Course 4
Natural Language Processing

What is Natural Language Processing? (pg. 17)

How does NLP work? (pg. 18)
Applications of NLP (pg. 19)
The future is Chatbots (pg. 20)

21 Course 5
Robotic Process Automation

What is robotic process automation? (pg. 22)

Benefits and risks of RPA (pg. 23)

iii
 Computer Vision

PART

01 Welcome to the first part of the course!

In this module you’ll discover In this part of the course we’ll
that computer vision (CV) is a revise and explore the nuts and
powerful technology that has a bolts of this technology, the two
variety of business applications. types of computer vision (classic
This versatility-of-use makes CV and deep learning), learn why CV
a articularly exciting and fruitful has been so disruptive to so
area of artificial intelligence for many industries (illustrated with
multiple industries. examples), and computer
vision’s potential applications.
We will round off our analysis
with ten success stories from
companies that apply CV today.

1
What is computer vision?

CV is a field of artificial Due to the increased capturing and processing

intelligence that trains machines capabilities of computers, CV now means big bucks
to understand and interpret the for businesses.
visual world. Appliances that use
or manipulate images tend to use Think about it:
CV. This may sound futuristic, but We process about 90% of data in visual form, so
CV has been around since the computer vision can tackle all the jobs that rely on
1950s! The barcode scanner is human vision – from traffic inspectors to MRI
one early example of CV in use. scan-analysis experts, sports umpires to checkout
assistants, truck drivers to equipment inspectors.
Thanks to the technological
advancements of the 21st century, By automating these tasks, computer vision can
computer vision’s applications process them faster and with higher accuracy than
have entered many facets of our human workers.
lives. As with many things we
consume daily, we can take CV for
granted. It is found in applications
as varied as food sorting, image
searches, and offsite monitoring.
When your camera identifies a
face and auto-focuses on it; when Businesses can, therefore,
Facebook auto-suggests whom
implement CV to increase
to tag in a photo; even when a
car drives automatically – these efficiency and cut costs,
are all examples of CV in action.
freeing up their teams to carry
out more creative tasks.

Computer
Real World Images Data
Vision

2
How does computer vision work?

Two types of CV exist: Classic and deep learning.

Classic CV Deep Learning CV

relies on pre-built libraries of requires neural networks to function, specifically
features. It collects images, convolutional neural networks (CNN). The key
labels them according to similar difference between the two types is that classic CV
characteristics, and groups them uses features that we have already input into the
in a dataset, or library of features. library, while deep learning CV generates its own
library of characteristics by using its CNN.

Classic Computer Vision

INPUT Feature Engineering Features Classifier with

(Manual Extraction shallow structure
+Selection)

OUTPUT

Deep Learning Computer Vision

INPUT Feature Learning + Classifier OUTPUT

(End-to-End Learning)

Computer vision can automate any task that requires the visual analysis of data,
from tomato sorting to driving a car. In this course we’ll use the former example to
illustrate how easy it is to innovate with CV.

3
Applications of computer vision

Image classification
groups visual files into categories. It helps us to identify and categorize objects.
E.g., pictures of dogs and cats; images of defective and working parts.

CAT CAT, DOG CAT, DOG

Single Object Multiple Objects

Image segmentation Object detection

partitions an image into helps to identify important parts of
multiple regions which can an image. Advanced object detection
be examined/ manipulated can recognize multiple objects in an
separately. E.g., Skype can image and draw identification boxes
now blur out the background around them. E.g., identifying human
behind a person on a call. life; oil spills.

Object tracking Object tracking

monitors movement – a Tesla car enables us to generate 3D-renders
was recently able to apply breaks from 2D-images. E.g., Google Maps;
on a highway seconds before a transferring clothing patterns.
collision happened ahead of it.
E.g., tracking pedestrian routes;
monitoring car speeds.

Edge detection Facial recognition

finds the outside edges of an identifies specific individuals.
object to identify image content E.g., automating the process
accurately. E.g., night vision. of photo tagging.

4
Face detection
facilitates applications to recognize facial features. E.g., camera focus;
identifying the number of faces in an image.

Brianna

Amy

Viviane

Optical character Pattern detection

recognition (OCR) matches repeated shapes, colors
recognizes words and numbers and other visual indicators in
in scanned printed and hand- images. E.g., handwriting
written documents, with recognition, fingerprint analysis.
accuracy rates of over 99%. E.g.,
manuscript study; note-taking.

Feature matching
finds similar characteristics and
groups them together. E.g., similar
objects in photographs.

Real-world use cases

At this point we will look at ten exciting, real world applications of CV

from industry giants such as eBay and Tesla!

5
 Deep Learning

PART

02 Welcome to the part two of the course!

Deep learning (DL) is an In this part of the course, you will
extremely powerful exponential get a bird’s eye view of the field
technology which is used in of DL understand how it works
applications ranging from and why it’s one of the most
cancer detection and mineral amazing fields of research which
exploration to productivity can put your company lightyears
tracking and fighting money ahead of the competition. This
laundering, saving companies part of the course will cover the
millions of dollars and ultimate essentials for you to get
changing lives. up-to-speed with DL and see
tangible examples of how this
technology can add substantial
value to your business.

6
Applications of computer vision

DL is a family of machine learning methods based on artificial neural networks.

There are two parts to this :

Machine learning So, what sets DL So, what is a DNN?

is defined as a discipline within apart from standard It’s an artificial construct designed
AI that teaches computers how methods of analysis to mimic how the human brain
to make predictions based on or even from ML? works. The way to think about it is
data. that a DNN will learn about your
The main difference is that in DL
data set just as a baby will learn a
there is no pre-defined
The human brain is the most language or how to differentiate
framework. In the case of
sophisticated system for data between objects.
non-DL methods there are
analysis known to us, it is
certain frameworks that we tried
therefore understandable that It has a brain, it has no predefined
to fit on the data to explain the
we wanted to create something network that we put in there and
patterns that we are seeing. If
similar inside our computers. simply by walking around and
they work then great, if they
interacting with humans, it
don’t – then we try something
Computer scientists have found automatically and slowly learns
else, until we find something
a way to mimic the human brain how to talk, how to walk, and how
that fits the data well and helps
inside machines: Artificial neural to do other things. So that’s how
us extract those insights.
networks. The goal of using these a DNN works as well – we try to
neural networks is to approach mimic the human brain. And that
In the case of DL, there is no
and solve general and complex is the beauty of DL; without
such thing. All we have is a Deep
problems in a similar way to how having to put any framework on
Neural Network (DNN) with
a human brain does. our data, we allow it to learn on
multiple layers.
its own.

DEEP LEARNING
The drawback with DL is that it requires a lot of data,
much more data than other algorithms. It might
CAR
require thousands, hundreds of thousands, or even NOT CAR
millions of examples before it can come up with
those features and truly learn. On the flip side, the
Input Feature Extraction Output
benefits that we get are incredible. + Classification

7
Building Neural Networks

If DL algorithms learn on their own, then why do we need

data scientists or ML engineers to build them?
And why are experts in this space so expensive?

Here we see a basic input-output type of neural Each one of these neural networks – whether it has
network: It has one input layer with 4 neurons zero hidden layers, or one, or two or 10 – is a valid
and one output layer with 1 neuron. Now what neural network. Moreover, we can also change the
we can do is add another layer in-between the number of neurons in each layer. For example, the
input and output layers. This layer is called a first hidden layer might have 8 neurons, while the
“hidden” layer and in this case it has 6 neurons. 2nd hidden layer might have only 2 creating and
We can add another hidden layer, and another. then back to 6 in the 3rd hidden layer – forming a
sort of a bottle-neck shape. That is also a valid
neural network.

These different ways that neural networks can be constructed are called architectures.

INPUT LAYER HIDDEN LAYERS OUTPUT LAYER

Input 1 hidden 1 hidden 2

Input 2 output 1

Input 3

Input n

Every purpose, every application, every challenge will have a different architecture that
serves it best. Finding the right neural network architecture is actually a very creative
process. And that’s why expert ML engineers and data scientists who build DL algorithms
are in such high demand and are so expensive to have on your team – because they are
doing highly creative work which requires their unique touch and input.

And once a suitable architecture is set up, the algorithm will do

the rest of the work itself.

8
Deep Learning and other areas of AI

Other areas of AI such as Reinforcement Learning,

Natural Language Processing and Computer Vision
represent some of the most disruptive exponential
technologies currently in existence.

What happens when we combine them

with DL?

We can see that while these technologies exist

within the domain of AI, they don’t fully fall into DL.
And this in itself, is already a very important point – However, when we do add DL to these
that Computer Vision, Natural Language technologies, what that means is that we’re
Processing and Reinforcement Learning don’t essentially adding the power of deep neural
always require DL in order to function. networks to those technologies:

Deep Natural Language Processing.

1 Reinforcement Learning becomes
Deep Reinforcement Learning.

DOCUMENTS PREPROCESSING DENSE

2 Computer Vision becomes Deep
Computer Vision.
EMBEDDING

OUTPUT OUTPUT UNITS HIDDEN

3 Natural Language Processing
becomes Deep Natural Language
LAYERS Processing.

As usual with neural networks, this approach has

the drawback of more costly computation. However,
if the computing power is available, then more often
than not, adding a deep neural network can
substantially improve accuracy of the algorithm.
Simply because now it has additional room to think
and experiment with different neural pathways.

You can often add DL to another

exponential technology to enhance it.

9
Real-world use cases

When attempting to develop an understanding of new technologies and

systems, it is imperative that we look beyond the hypothetical and examine
some real-world examples.

At this stage we will look at examples of how various industries have

incorporated DL into their corporate life.

You will gauge how global companies have used DL in practice to increase
productivity, profit, compliance and security across diverse sectors.

10
 Reinforcement
Learning

PART

03 Welcome to the part three of the course!

Reinforcement Learning (RL) is an In this module, you will get
extremely powerful exponential a bird’s eye view of the field
technology which has helped of RL, understand how it works
companies gain the upper hand and why it’s one of the most
in a huge variety of situations – disruptive technologies which
from Google cutting down their can take operations, marketing
electricity bill by 40%, to Alibaba and even innovation to the next
increasing their advertising ROI level.
by 240%!

11
What is Reinforcement Learning?

There are three main groups of algorithms in ML: Unsupervised Learning,

Supervised Learning and Reinforce RL.

It can be valuable to appreciate the differences:

Unsupervised Learning Reinforcement Learning (RL)

is used for discovery of new patterns. is something entirely different. RL is an area of ML
where an imaginary agent is being presented with
For example, clustering of customers into groups a problem and is being rewarded with a “+1” for
based on their similarities. The core principle here finding a solution to the problem or punished with
is that the resulting groups did not exist prior, but a “-1” for not finding a solution.
rather are suggested by the machine in the
process. Unlike with Supervised Learning, the agent is not
given instructions on how to perform the task.
Instead, it performs random actions and interacts
Supervised Learning with its environment. It learns through trial and error
is when we teach a machine to search and identify which actions are good and which actions are bad.
patterns that we have seen before.
environment
For example, classification of pictures of dogs and
cats into the two categories “dogs” and “cats”. First,
interpreter
we show the algorithm thousands of already
reward action
labelled images so it can extract features that are
essential to dogs and features that are essential to
cats. After this the algorithm will be able to agent
categorise new images as either of “dogs” or of state
“cats”. The difference of this approach to
unsupervised learning is that we have to first
provide the labelled data for the algorithm to learn.

Reinforcement Learning in Humans

RL is surprisingly similar to real life and is based on the same principles

as a baby learning to walk, and because of that – because it is so similar
to the way our own intelligence works – out of all existing algorithms,
RL is the closest we have gotten to true AI.

12
Advantages of Reinforcement Learning

Since Supervised
Learning relies on
large labelled data- That is the number one The fourth advantage: On-line
sets, RL has a much advantage of RL: It doesn’t learning.
require large labelled datasets.
higher scope of RL runs in real-time. RL combines
application than any This is a huge advantage exploration (when the machine
because, as the amount of data tests new approaches on the fly
form of Supervised
in the world grows, it is to find better solutions) and
Learning. becoming ever more costly to exploitation (when the machine
label it for all required exploits the best solutions which
applications it has already found). This means
that it can bring results while
improving at the same time.
Other algorithms would require
The second advantage: RL is re-training and re-deployment.
innovative. RL just keeps going.

Supervised Learning is where

the algorithm learns to imitate
whoever provided the data set. The fifth advantage: RL is
The algorithm can learn to do goal-oriented.
the task as well as (or better
than) the teacher but can never RL can be used for a sequence of
learn a completely new actions.
approach to solving the While Supervised Learning is
problem. On the other hand, RL mostly used in an input->output
algorithms can come up with manner, RL can be used for tasks
entirely new solutions that were with objectives – such as robots
never even considered by playing soccer or self-driving cars
humans. getting to their destinations.

The third advantage: RL is

bias-resistant.

If there’s bias in the way the data

is labelled, Supervised Learning
algorithms will pick it up. In this
sense, RL algorithms are better
tools to find solutions free from
bias or discrimination.

12
Reinforcement Learning in Marketing

Because of its “on-line learning” Virtually any business could benefit from cutting-edge
advantage – the ability to auto-correct tech in its marketing and that’s why in this section of
on the fly – RL’s most popular current the course we will look into 5 examples of how RL will
business application is, perhaps, revolutionize the way we do marketing and then further
within marketing. examples of how companies are already using RL.

RL in Marketing : Creating Personalized Recommendation

ACTION

PAGE REQUEST

ENGINE

13
 Natural Language
Processing

PART

04 Welcome to the part four of the course!

Natural Language Processing In part four of this executive
(NLP) is an extremely powerful briefing course you will get an
and potent platform which is enterprising perspective of NLP,
used in applications ranging understand how it works and
from virtual personal assistants how it can be used to leverage AI
and automatic translations to and provide immense value to
legal document reviews and your business.
chatbots - all of which save
companies millions of dollars.

14
Building
Advantages
Neural
of Reinforcement
Networks Learning

NLP is the field of AI

concerned with the Computer
interactions between Science

computers and human

(natural) languages, in
NLP
particular how to
Artificial Human
program computers to Intelligence Language
process and analyze
large amounts of
natural language data.
But before we dive into NLP we need to understand the two types
of data that exist: Structured and unstructured.

Structured data
is data that has a pre-defined specific format. For example,
spreadsheets, tables, databases.

Unstructured data,
on the other hand, is data that This data, like all data, contains
doesn’t follow a pre-defined power: Power to know your
specific format. For example, customers better, power to
emails, chats, blogs, books, audio
increase efficiency and the power
files, video files, images. We to scale. However, because of its
humans communicate unstructured format, it is much
predominantly in unstructured harder to leverage. And that is
data, specifically via natural why NLP is an exploding field
language. right now. Businesses are in a
race to be the first in their
industry to unlock the potential of
their unstructured data. In the
And that’s why, based on
following tutorials we will explore
various sources, unstructured
what opportunities might exist in
data accounts for 70-90% of data
your business to leverage NLP.
in the world. Most of this data
contains text in some form –
whether written or audio.

15
How does Natural Language Processing work?

There are Two parts

to NLP:
Natural Language
Understanding (NLU) Natural Language Natural Language
Understanding Generation
and Natural Language refers to mapping the given input is the process of producing
Generation (NLG). from Natural Language into formal meaningful phrases and
representation and analysing it. sentences in the form of natural
language from some internal
If the input is in the form of audio, representation. NLG is generally
then Speech Recognition is applied much easier than NLU since
first to convert it into text. Then the when we know the meaning that
hard part starts – interpreting the needs to be expressed, there are
meaning from the text. Human certain rules of language such as
language is complex, some words, syntax and semantics that need
for example, “bank” or “leaves” can to be followed in order to create
carry different meaning depending an appropriate sentence.
on the context they are in.
If the text needs to be put into
Once the meaning is extracted audio, like in the case of Siri or
categorise the input and come up Alexa, then Speech Generation
with an appropriate action or comes in, but this is not always
response. required.

This is where NLG, or Natural

Language Generation, may
ACTION
be required.

Natural Language Processing

Natural Language Understanding

Natural Language Generation
(Linguistic)

Natural Language Text Phonology

Syntax

Pragmatics
Semantics

Morphology

17
Applications of Natural Language Processing

Every industry can leverage unstructured data in many different ways. In this part of the course we will look
at some of the possible applications of NLP to help you innovate and come up with ideas for your business.

1 Sentiment analysis to help you understand how users feel about your company or product

2 Speech recognition to convert audio into text for further processing.

3 Chatbots to provide 24/7/365 customer support.

Machine
4 translation
to help you reach new markets with minimal investment.

similar to what Gmail does to help you increase your employee’s efficiency
Autocompleting
5 text
through a centralized knowledge base. Or to improve how your customers
interact with your product / website.

to make sure that documents don’t contain errors. Especially relevant

6 Spell checking in industries with high compliance requirements such as banking,
insurance and finance.

to locate relevant information faster across all data facilities,

7 Keyword search leading to increased efficiency.

Advertisement For example, when you search for a new car on Google or write an
8 matching email about it to a friend, you will later receive car adverts.

Information
9 extraction
to extract the essence from large volumes of unstructured data.

10 Spam detection to clean your email inbox.

to create new text, e.g. client contracts, research documents, training

11 Text generation materials, and so on.

18
 Automatic
12 summarization
to add summaries to existing documents.

Questions
13 answering
to answer specific questions based on disparate information sources.

14 Image captioning to annotate images.

Videos can take a long time to watch. Sometimes it’s faster to extract
15 Video captioning information from them in the form of text and then process it further
using other NLP techniques like summarization.

Applications of Natural Language Processing

All NLP applications are Over 250 billion customer support requests are made
every year, and it costs businesses an incredible 1.3
growing fast. However, one of trillion dollars to service them. Most of these requests
them is truly skyrocketing at are simple and can be answered by AI, as long as it
has a deep knowledge base and is good enough at
astronomical rates in a huge understanding natural language.
variety of industries… we’re
talking about chatbots!

19
 Robotic Process
Automation

PART

05 Welcome to the final part of the course!t

Welcome to the final part of the In this section we’ll revise and
course! As you’ll discover, robotic explore the nuts and bolts of this
process automation (RPA) is resource, the risks and benefits
a powerful and emerging field of introducing RPA to your
that has already demonstrated company. We will round off our
a variety of real-world applications. analysis with ten success stories
The potential for improved from companies that apply RPA.
quality, increased savings and
more efficient workflows mean With these informational tools,
that this is a vital growth area for this video part of the course will
businesses. show you just how easy it is to
implement robotic process
automation into your business.

20
Deep Learning and other areas of AI

RPA is the use of easily programmable software to

handle high-volume, repeatable tasks that previously
were manually performed by humans.

Robots are automating blue-collar jobs. For example,

they are beginning to replace truck drivers, factory
workers and many other labour-intensive roles.
Increasingly, RPA is also automating white-collar jobs.
Many companies are pivoting from people to software One of the reasons that RPA
for high-volume highly-transactional manual processes.
is proving increasingly
popular is the fact that it is
more financially beneficial
to companies.

On average, it provides
25-40% cost savings,
compared to 5-10% cost
It is essential to have a good grounding of what RPA savings of using traditional
is in order to understand its growing prevalence. We
will go through these in this part of the course. outsourcing business models.

Despite its growth, RPA is sometimes misunderstood

and confusion may arise between RPA and Intelligent
Automation (IA). The difference lies in the fact that RPA
is the application of technology which enables
computer software configuration to partially or fully
automate human activities which are manual, repetitive
or rulebased. On the other hand, intelligent automation
intends to mimic human behavior, such as perceiving,
gathering evidence or reasoning. It is for this reason that
IA is best for processes which involve unstructured data
from non-standard sources.

RPA has developed slowly over the past decade and is

now widely used due to its ability to give returns quickly
and cost effectively. IA is enabled by and inherent to
cognitive technologies and has huge transformative
potential in the near future.

The increased use of RPA is having a marked impact

on many sectors.

21
Benefits and risks of RPA

RPA has a number of benefits,

hence it’s growth.

However it also has some drawbacks.

In this part of the course, we will explore both!

Afterwards we will: examine some real-world examples

of how various industries have incorporated RPA into
their corporate life. Although some remain unnamed,
you will gauge how larger companies and more
comprehensive rollouts of RPA tend to result in
increased savings and benefits.

22
Hungry for more?
Find us at www.superdatascience.com,
where you can join us on our other courses
that deconstruct exponential technologies
in under 30 minutes!