Deep learning

M A C H I N E L E A R N I N G F O R E V E R YO N E

Sara Billen
Curriculum Manager, DataCamp
What is deep learning?
AKA: Neural Networks
Basic unit: neurons (nodes)

Special area of Machine Learning

Requires more data

Best when inputs that are images or text

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue
 

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Predicting box of ce revenue

MACHINE LEARNING FOR EVERYONE

Deep learning
  Neural networks are much larger

Deep learning: neural network with many

neurons

Can solve complex problems

MACHINE LEARNING FOR EVERYONE

When to use deep learning?
Lots of data

Access to processing power

Lack of domain knowledge

Complex problems
Computer vision

Natural language processing

MACHINE LEARNING FOR EVERYONE

 Let's practice!
M A C H I N E L E A R N I N G F O R E V E R YO N E
 The process
M A C H I N E L E A R N I N G F O R E V E R YO N E

Sara Billen
Curriculum Manager, DataCamp
Computer vision
Helps computers see and understand the content of digital images

MACHINE LEARNING FOR EVERYONE

Image data

MACHINE LEARNING FOR EVERYONE

Image data

MACHINE LEARNING FOR EVERYONE

MACHINE LEARNING FOR EVERYONE
MACHINE LEARNING FOR EVERYONE
MACHINE LEARNING FOR EVERYONE
MACHINE LEARNING FOR EVERYONE
Training the neural network

MACHINE LEARNING FOR EVERYONE

Applications
Facial recognition

Self-driving vehicles

Automatic detection of tumors in CT scans

Deep fake

...

MACHINE LEARNING FOR EVERYONE

 Let's practice!
M A C H I N E L E A R N I N G F O R E V E R YO N E
 Natural Language
Processing
M A C H I N E L E A R N I N G F O R E V E R YO N E

Sara Billen
Curriculum Manager at DataCamp
Natural Language Processing (NLP)
The ability for computers to understand the meaning of human language

MACHINE LEARNING FOR EVERYONE

Bag of words

MACHINE LEARNING FOR EVERYONE

Bag of words
"U2 is a great band" "Queen is a great band"

Word Count Word Count

U2 1 U2 0

Queen 0 Queen 1

is 1 is 1

a 1 a 1

great 1 great 1

band 1 band 1

MACHINE LEARNING FOR EVERYONE

Bag of words: n-grams
"That book is not great" 2-gram (bi-gram)

Word Count Word Count

That 1 That book 1

book 1 book is 1

is 1 is not 1

not 1 not great 1

great 1

MACHINE LEARNING FOR EVERYONE

Bag of words: limitations
Word counts don't help us consider synonyms

Example: "blue"
"sky-blue"

"aqua"

"cerulean"

Want to group as a single feature

MACHINE LEARNING FOR EVERYONE

Word embeddings
Word embeddings
Create features that group similar words

Features have a mathematical meaning:

king - man + woman = queen

MACHINE LEARNING FOR EVERYONE

Language translation

MACHINE LEARNING FOR EVERYONE

Applications
Language translation

Chatbots

Personal assistants

Sentiment analysis

...

MACHINE LEARNING FOR EVERYONE

Deep learning
Two types of problems
Computer vision

Natural language processing

Why deep learning?

Complex problems

Automatic feature extraction

Lots of data

MACHINE LEARNING FOR EVERYONE

 Let's practice!
M A C H I N E L E A R N I N G F O R E V E R YO N E
 Limits of machine
learning
M A C H I N E L E A R N I N G F O R E V E R YO N E

Sara Billen
Curriculum Manager, DataCamp
Data quality

Garbage in garbage out

Output quality depends on input quality

MACHINE LEARNING FOR EVERYONE

How it can go horribly wrong
Amazon's gender-biased recruiting tool Recruiting software to help review resumes

Preferred men because it learned from historic

data when more men were hired

It downgraded resumes that

contain the word "women"

implied the applicant was female

MACHINE LEARNING FOR EVERYONE

How it can go horribly wrong
Microsoft's AI chatbot

MACHINE LEARNING FOR EVERYONE

Beware
Don't blindly trust your model

Awareness is key

Pay attention to your data

A machine learning model is only as good as the

data you give it

MACHINE LEARNING FOR EVERYONE

Quality assurance
High-quality data requires:
Data analysis

Review of outliers

Domain expertise

Documentation

MACHINE LEARNING FOR EVERYONE

Explainability

MACHINE LEARNING FOR EVERYONE

Explainability

Transparency to increase trust, clarity, and understanding

Use cases: business adoption, regulatory oversight, minimizing bias

MACHINE LEARNING FOR EVERYONE

Explainable AI
Black box Explainable AI
Deep learning Traditional machine learning

Better for "What?" Better for "Why?"

Highly accurate predictions Understandable by humans

MACHINE LEARNING FOR EVERYONE

Example: Explainable AI

1. Prediction: Will the patient get diabetes?

2. Inference: Why will this happen

MACHINE LEARNING FOR EVERYONE

Example: Inexplicable AI

Prediction only: Which letter is this likely to be?

MACHINE LEARNING FOR EVERYONE

 Let's practice!
M A C H I N E L E A R N I N G F O R E V E R YO N E
 Congratulations!
M A C H I N E L E A R N I N G F O R E V E R YO N E

Lis Sulmont
Curriculum Manager, DataCamp
Chapter 1
What is machine learning?

Machine learning concepts and work ow

MACHINE LEARNING FOR EVERYONE

Chapter 2
Different types of machine learning

How we evaluate and improve machine learning models

MACHINE LEARNING FOR EVERYONE

Chapter 3
Deep learning, including computer vision and natural language processing

Limits of machine learning

MACHINE LEARNING FOR EVERYONE

What's next?

MACHINE LEARNING FOR EVERYONE

What's next?

MACHINE LEARNING FOR EVERYONE

 Congrats!
M A C H I N E L E A R N I N G F O R E V E R YO N E