Sei sulla pagina 1di 31

Dr.

Gustavo Snchez (gsanchez)


Universidad Simn Bolvar - Venezuela

INTRODUCTION TO
MACHINE LEARNING

Outline
General

Concepts
Decision Trees
Neural Networks I
Neural Networks II
Evolutionary Computation I
Evolutionary Computation II
Case Study I
Case Study II

What is Learning?
Process

of

gaining
knowledge or
skill by studying,
practicing, being
taught, or
experiencing
something.

What is Learning?

What is Learning?
Is

learning equal
to imitating?

What is Knowledge?
Stored

information or
models used to
interpret, predict
and appropriate
respond to the
outside world

What is Knowledge?

What about Intelligence?


Ability

to meet
(novel)
situations
successfully by
proper behavior
adjustments

Artificial Intelligence
The

science of
making
computers do
things that
require
intelligence
when done by
humans.
Alan Turing (1912-1954)

What is Machine Learning?


Discipline

concerned with
the design and
implementation
of computer
systems that can
learn
autonomously

What is Machine Learning?

Function Approximation
An

important problem
in Machine Learning
is inferring a
functional
relationship between
a set of attribute
variables and
associated response
variables

Attribute Variables

Functional
Relationship

Response Variables

Function Approximation
a,b

Functional
Relationship

Memory, algorithm or both?

a*b

Function Approximation
We

may wish to build


a model that can
perform an automatic
failure diagnosis from
a data set comprised
of measurements

Measurements

Model

Failure diagnosis

Pattern Recognition/Classification
Humans

are good at
pattern recognition!

Pattern Recognition/Classification
Face,

voice and even


smell and taste

Pattern Recognition/Classification
We

are able to
recognize patterns,
even after significant
changes

Digital Image Processing

Information must be converted into a digital


format before it can be stored or processed
on a computer

Pattern Recognition/Classification
It

can be
defined as the
process
whereby a
received
pattern/signal
is assigned to
a prescribed
number of
classes
(categories)

Pattern Recognition/Classification
During

the
training the
learning
system is
presented a set
of input
patterns and
their categories

Pattern Recognition/Classification
Then

a new
pattern is
presented, not
seen before,
and the system
is expected to
identify its
class

Pattern Recognition/Classification
A

pattern is
represented by
a set of m
observables.
Feature
extraction is
described by a
transformation
into a qdimensional
space (q < m)

Pattern Recognition/Classification
Classification

is

a
transformation
that maps the
feature space
into one of the
classes in a rdimensional
space

Pattern Recognition/Classification
A

twodimensional
feature space:
point
represents an
individual who
is 6ft tall and
weighs 200lb

Pattern Recognition/Classification

Graphically, in
this example
we have
patterns in the
Feature Space
(points in R2)
belonging to
two categories
or classes

Feature Space

Learning Process

Supervised
Learning: A teacher
has the knowledge,
represented by a
set of input-output
examples

Learning Process

Unupervised
Learning: No
teacher to oversee
the learning
process.

The system must


be able to generate
its own response

Statistical Learning Process

Consider a stochastic phenomenon


described by a random vector X of
independent variables and a random scalar
D representing a dependent variable

We

have N realizations of X and D,


denoted:

1
Training Sample

Statistical Learning Process

We do not know the functional relationship


between X and D, so we propose a
regressive model:

= +

where
is a random expectational error
that represents our ignorance about the
dependance of D and X

Statistical Learning Process

When the mean value of

given any

realization is zero
have the following properties:

= 0 we

[ ] =
The

regression function is the conditional


mean of the model output!

Statistical Learning Process

The expectational error is uncorrelated with


the regression function!

[ ] = 0

Potrebbero piacerti anche