Scribd
Upload
10 views21 pages

W9 ML Overview NRG

Uploaded by

Zhaocun Sun
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
10 views21 pages

W9 ML Overview NRG

Uploaded by

Zhaocun Sun
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Machine Learning

Comp2208: Intelligent Systems

Dr. Nery Riquelme-Granada

Vision, Learning & Control Research Group


Lecture Outcomes

By the end of this lecture, you will:

• Understand the relationship between Machine Learning


(ML) and Artificial Intelligence (AI)
• Be able to define Machine Learning
• Differentiate among the various types of learning
• Explore examples of ML research and applications
• Use standard terminology in Machine Learning
• (Hopefully) Feel excited about the potential of ML

1
The story so far

2
Artificial Intelligence and Machine Learning

• Classification is a type of Machine Learning problem, for


which modern methods, e.g. Deep Learning, have
achieved remarkable results
• Machine Learning is a fundamental component of the
Artificial Intelligence enterprise

3
Figure 1: Plot by Zhang et. al.
What is Machine learning?

Arthur Samuel (1959):


Machine Learning is the field of study that gives the
computer the ability to learn without being explicitly
programmed

Figure 2: Original Article

4
What is Machine learning?

Arthur Samuel (1959):


Machine Learning is the field of study that gives the
computer the ability to learn without being explicitly
programmed

• Moves in the game


should not be
explicitly coded
• Static programs fall
short for tackling
this problem

5
What is Machine Learning?

Tom Mitchell (1997):


A computer program is said to learn from experience E
with respect to some class of tasks T and performance
metric P, if its performance of tasks in T , as measured
by P, improves with experience E .

6
What is Machine Learning?

• T : Playing
Checkers
• P: Percent of
games won against
a human adversary
• E : Games played
against itself

7
Types of Learning

Tasks (T ), Measures of Performance (P) and Experience (E )


roughly define three types of Learning:
• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning

8
Types of Learning

Tasks (T ), Measures of Performance (P) and Experience (E )


roughly define three types of Learning:
• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning

9
Supervised Learning

• Training data: {(x1 , y1 ), (x2 , y2 ), (x3 , y3 )...(xn , yn )}


• y1 is the label for feature vector x1

Classification
labels are discrete values, e.g. {spam, no-spam}, {cancer,
healthy}, {Setosa, Versicolour, Virginica}

10
Supervised Learning

• Training data: {(x1 , y1 ), (x2 , y2 ), (x3 , y3 )...(xn , yn )}


• y1 is the label for feature vector x1

Regression
Labels are continuous values: stock market, blood pressure
measurements, anything expressed as a Real Number, i.e.
yi ∈ R

10
Example of training data

11
Anatomy of a feature vector

 
 sun 
 
 hot 
 
x1 = 
 
 high 

 
 
weak
• each component of the vector is a feature or
measurement.

12
Algorithms for classification

• Classifiers find a function h(x) → y


• Standard success metric, a.k.a P: fraction of correctly
classified
Two well-rounded methods

Figure 3: K Nearest Neighbors Figure 4: Decision Trees


13
Others: Logistic Regression, Neural Networks, SVM,
Unsupervised Learning

• Input data: {x1 , x2 , x3 ...xn }


• No label information available - only feature vectors
• Classic problem: Clustering
• Applications: Anomaly detection, Knowledge discovery,
etc.

14
Handwritten recognition learning problem

• T : Recognising and
classifying
handwritten words
within images
• P: Percent of words
correctly classified
• E : Database of
handwritten words
with given
Figure 5: Source: A Database for
classifications
Handwritten Text Recognition Research

15
Self-driving car learning problem
• T : driving on a
public four-lane
highway using vision
sensors
• P: Average distance
travelled before an
error (judged by a
human supervisor)
Figure 6: Source: Self-driving cars: A
• E : Sequence of
survey
images and steering
commands recorded
while observing a
human driver
16
Gene similarities learning problem
• T : Group gene
expressions of
individuals based on
similarity
• P: Average number
of individuals
correctly grouped Figure 7: Source: Identifying regulatory
together mechanisms using individual variation
• E : set of gene reveals key role for chromatin
expression profiles modification
and genetic markers
from a population

17
Why we need Machine Learning?

Tasks That Are Too Complex to Program


• Computer vision: we know to detect objects but have no
idea how we do it
• Search engines: a human can’t read the entire internet

Adaptivity
• speedy and adaptive development of programs

18
Lecture summary

• Machine Learning is about turning experience into


expertise.
• Tasks, performance, and experience are high-level
descriptions of ML problems.
• There are three ML settings: supervised, unsupervised,
and reinforcement learning.
• Supervised learning problems can be classified as either
classification or regression.
• Classification is a supervised learning task where we
predict discrete labels for given objects.
• K-Nearest Neighbors (KNN) and Decision Trees are
algorithms used for classification problems.
• Machine Learning enables computer programs to be
19
adaptive and tackle complex tasks.

You might also like