Module 1: What is Machine
Learning
Module 1 Objectives:
At the conclusion of this week, you should be able to:
1) Describe what machine learning is and does
2) Explain why we should care about machine learning
3) Identify the common types of machine learning tasks
4) Define common ML terms to be able to understand
articles and conversations about ML
Introduction to Machine
Learning
What is Machine Learning?
“Field of study that gives computers
the ability to learn without being
explicitly programmed” – Arthur
Samuel, IBM, 1959
Instead of providing a computer with
exact instructions to solve a problem,
we show it examples of the problem to *"! $"&$( "'% (!("$
&&#% " "!%( "$ ( !)# #'$
solve and let it figure out how to solve
it itself
ML vs. traditional software
" "
$
!
AI vs. Machine Learning
!
&&#% " "!%) "$)
%%' " #
Brief History of AI/ML
Machine Learning Today
Explosion in data
Ubiquitous internet connectivity
Advances in sensor technology
Smart connected devices
Deep learning has made what was impossible,
possible
Massive increase in computational power – GPUs
Huge sets of labeled data for training
Algorithmic advances
Pervasiveness of machine learning models in
products and systems we interact with daily
Where Do We Find ML?
Where Do We Find ML?
Data Terminology
Data Comes in Many Forms
Data Comes in Many Forms
Almost anything can be turned
into numbers:
Measurements
Text
Images
Sound
Video
Data may have different
relationships:
Spatial relationships
Temporal relationships
Structured vs. Unstructured Data
Structured data Unstructured data
Set structure based on pre-defined Does not follow a defined format of
fields for each record fields
Often stored in relational databases Many types – images, videos,
Easy to enter, search and analyze sounds, text
Works well with common tools Requires specialized tools to work
with
)
& )
Continuous vs Categorical Data
! %$#"
Time series data
Series of data points organized in time order
Points are usually equally spaced by time
Assumptions:
Time is considered one-way #
Points close together in time are more related than points further apart
Terminology
!" " $
" # & $%#
! ! # " #
What is a Model?
What is a model?
A model is an approximation of the
relationship between two variables
!
! ' ( ! ' (
# $
What is a model?
!
!
What is a model?
What is a model?
!
!
Building a model
To create a model we define four things:
1. Features to use
2. Algorithm – acts as a form/template for model
3. Hyperparameter values for algorithm
4. Loss function to optimize
We train our model using historical data:
Algorithm & hyperparameters provide overall
model form
“Learn” values for the model which minimize
loss function
Types of Machine Learning
Types of Machine Learning
% $ !
!
!
$
& $
$
# $ $ !
! $
Supervised vs. Unsupervised Learning
%"'$ &&#% )))$%$&!& '$'#$( % $! !!'!%'#$( %
$! !'#$( % $! !'%%!!"&& "!
Supervised vs. Unsupervised Learning
%"'$ &&#% )))$%$&!& '$'#$( % $! !!'!%'#$( %
$! !'#$( % $! !'%%!!"&& "!
Regression vs. Classification
Predict one or more numerical target Predicts a class / category – either
variables binary or out of a set
E.g. home price, number of power E.g. lung disease detection, identifying
outages, product demand types of plants, sentiment analysis,
detecting spam
What ML Can and Cannot
Do Well
What ML can do well*
Automate straightforward tasks
Make predictions by learning input-
output relationships
Personalize for individual users
* Given sufficient quantity and quality of data
What ML cannot do well
Understand context
Determine causation
Explain “why” things happen
Determine the impact of
interventions / find solutions
Wrap-up
Wrap Up
ML enables computers to learn from
experience without explicit instructions
Data is the key!
Types: supervised vs. unsupervised vs.
reinforcement learning
Useful for automation, prediction and
personalization
NOT useful for explaining “why” or “how
to fix”