Introduction to Machine Learning

advertisement
Neural Networks
and Machine Learning Applications
CSC 563
Prof. Mohamed Batouche
Computer Science Department
CCIS – King Saud University
Riyadh, Saudi Arabia
mbatouche@ccis.ksu.edu.sa
Syllabus

Course Description

Humans are capable of tackling extremely difficult problems
without the benefit of an a priori solution. They learn from
experience and can often transfer knowledge acquired to
novel instances or even whole new tasks.
“Are machines capable of similar problem solving
process?”

In this course, we will study multiple machine learning
models including neural networks, support vector machines,
Bayesian learning, reinforcement learning and learning
through emergent behaviour.
2
Syllabus
Recommended Books:

Introduction to Machine Learning, Eethem Alpaydin, The MIT Press,
October 2004.

Machine Learning, Tom Mitchell, McGraw Hill, 1997.

Introduction to Machine Learning, Nils J. Nilsson (available (free)
from the Web).

Artificial Intelligence Illuminated, BEN COPPIN, Jones and Bartlett
Illuminated Series, 2004

Research papers (available from the Web).
3
Syllabus

Grading

MT1
MT2
20%
20%

Final exam
40%

Research papers
20%


KSU Group: http://faculty.ksu.edu.sa/mohamedbatouche
4
Syllabus

Course Overview (main topics)







What is Machine Learning?
Neural Networks Learning
Support Vector Machines
Bayesian Learning
Reinforcement Learning
Artificial Life
Machine Learning applications
5
What is Machine Learning?
What is Learning ?
Herbert Simon: “Learning is any
process by which a system improves
performance from experience.”

What is the task?



Classification
Problem solving / planning / control
7
What is Learning ?
Classification: Assign object/event to
one of a given finite set of categories.






Medical diagnosis
Spam filtering in email
Financial investments
Spoken words
Handwritten letters
8
What is Learning ?
Problem Solving/Planning/Control:
Performing actions in an environment
in order to achieve a goal.





Playing checkers, chess, or backgammon
Driving a car or a jeep
Flying a plane, helicopter, or rocket
Controlling a mobile robot
9
What is Machine Learning ?

Definition: A computer program is said
to learn from experience E with respect
to some class of tasks T and
performance measure P, if its
performance at tasks in T, as measured
by P, improves with experience E.
10
Why Study Machine Learning?
AI began as an attempt to understand the nature of
intelligence, but it has grown into a scientific and
technological field affecting many aspects of commerce
and society. The main goals of AI & ML are:

Engineering better computing systems:
Develop systems that are too difficult/expensive to
construct manually because they require specific
detailed skills or knowledge tuned to a specific task
11
Why Study Machine Learning?

Cognitive Science:
Computational studies of learning may help us
understand learning in humans and other
biological organisms.
Even as AI Technology becomes integrated into the fabric
of everyday life. AI researchers remain focused on the grand
challenges of automating intelligence.
12
Examples of Successful
Applications of Machine
Learning :

Learning to recognize spoken words (Lee, 1989; Waibel,
1989).

Learning to drive an autonomous vehicle (Pomerleau,
1989).

Learning to classify new astronomical structures (Fayyad
et al., 1995).

Learning to play world-class backgammon (Tesauro 1992,
1995).
13
Why is Machine Learning
Important ?


Some tasks cannot be defined well,
except by examples (e.g., recognizing
people).
Human designers often produce
machines that do not work as well as
desired in the environments in which
they are used.
14
Why is Machine Learning
Important ?



The amount of knowledge available about
certain tasks might be too large for explicit
encoding by humans (e.g., medical
diagnostic).
Environments change over time.
New knowledge about tasks is constantly
being discovered by humans. It may be
difficult to continuously re-design systems “by
hand”.
15
ML Topics:
A Quick Introductory Overview
The main ML topics we’ll cover in this
course:

neural networks, Reinforcement Learning …

Artificial Life (Cellular Automata, GAs, GP …)
16
ML Topics:
A Quick Introductory Overview
Developing Intelligent Program Systems
Two main approaches:


Symbolic Approach
Expert systems: Feed the system with knowledge.
There is no learning !!
Connectionist Approach
Neural Networks: The system learns from
examples by itself.
17
Symbolic Approach
18
ML Topics:
A Quick Introductory Overview
Expert system


expert system (or knowledge-based system): a program
which encapsulates knowledge from some domain,
normally obtained from a human expert in that domain
components:





Knowledge base (KB): repository of rules, facts
(productions)
working memory: (if forward chaining used)
inference engine: the deduction system used to infer
results from user input and KB
user interface: interfaces with user
external control + monitoring: access external databases,
control,...
19
ML Topics:
A Quick Introductory Overview
Expert System

Why use expert systems:





commercial viability: whereas there may be only a few experts whose time
is expensive and rare, you can have many expert systems
expert systems can be used anywhere, anytime
expert systems can explain their line of reasoning
commercially beneficial: the first commercial product of AI
Weaknesses:

expert systems are as sound as their KB; errors in rules mean errors in
diagnoses

automatic error correction, learning is difficult

the extraction of knowledge from an expert, and encoding it into machineinferrable form is the most difficult part of expert system implementation
20
Connectionist Approach
21
ML Topics:
A Quick Introductory Overview
Machine Learning : Neural Nets


Artificial Neural Networks: Artificial Neural Networks are crude
attempts to model the highly massive parallel and distributed
processing we believe takes place in the brain.
Two main areas of activity:

Biological:
Try to model biological neural systems.

Computational: develop powerful applications.
22
ML Topics:
A Quick Introductory Overview
Machine Learning : Neural Nets
Neural nets can be used to answer the
following:

Pattern recognition: Does that
image contain a face?

Classification problems: Is this cell
defective?

Prediction: Given these symptoms,
the patient has disease X



Forecasting: predicting behavior
of stock market
Handwriting: is character recognized?
Optimization: Find the shortest
path for the TSP.
23
ML Topics:
A Quick Introductory Overview
Machine Learning : Neural Nets
Strength and Weaknesses of ANN

Examples may be described by a large number of attributes (e.g.,
pixels in an image).

Data may contain errors.

The time for training may be extremely long.

Evaluating the network for a new example is relatively fast.

Interpretability of the final hypothesis is not relevant (the NN is
treated as a black box).
24
ML Topics:
A Quick Introductory Overview
Modern AI:
Artificial Life, Complex Systems …
25
ML Topics:
A Quick Introductory Overview
Artificial Life & Complex Systems


Artificial Life: An attempt to better understand “real” life by
in-silico modeling of the entities we are aware of.
Motivations:


The Emergent properties in life motivate scientists to explore
the possibility of artificially creating life and expecting the
unexpected.
An Emergent property is created when something becomes
more than sum of its parts.
26
AI Topics:
A Quick Introductory Overview
Artificial Life : Cellular Automata
• The CA space is a lattice of cells (usually 1D, 2D, 3D)
with a particular geometry.
• Each cell contains a variable from a limited range of
values (e.g., 0 and 1).
• All cells update synchronously.
• All cells use the same updating rule,
depending only on local relations.
• Time advances in discrete steps.
27
AI Topics:
A Quick Introductory Overview
Cellular Automata:
Game of Life
Cellular Automata (CA) is an
array of N-dimensional ‘cells’ that
interact with their neighboring cells
according to a pre-determined set of
rules, to generate actions, which in
turn may trigger a new series of
reactions on itself or its neighbors.
The best known example is
Conway’s Life, which is a 2-state
2-D CA with simple rules (see on
right) applied to all cells
simultaneously to create generations
of cells from an initial pattern.
Conway’s Life: Rules
A living cell with 0-1 8-neighbors
dies of isolation
A living cell with 4+ 8-neighbors
dies from overcrowding
All other cells are unaffected
28
AI Topics:
A Quick Introductory Overview
Artificial Life : Genetic Algorithms
• Genetic Algorithms (GA) provide a learning method based on principles of
biological evolution.
• New hypotheses are generated by mutating and recombining current
hypotheses.
• At each step we have a population of hypotheses from which we select the
most fit.
• GA do a parallel search over different parts of the hypothesis space.
29
AI Topics:
A Quick Introductory Overview
Artificial Life : Genetic Programming
+
Genetic programming (GP)
Genetic programming is a form
of evolutionary computation in
which the individuals in the
population are computer
programs.
Programs are normally
represented by trees. A program
is executed
by parsing the tree.
Sin
Sqrt
x
+
^
x
y
2
F = sin(x) + sqrt( x^2 + y)
30
AI Topics:
A Quick Introductory Overview
Artificial Life : Genetic Programming
Genetic programming (GP)
Programming of Computers
by Means of Simulated Evolution
How to Program a Computer
Without Explicitly Telling It What
to Do?
31
What is Machine Learning ?

To conclude:


ML is a very fascinating field. It can help us solve difficult,
real-world problems, creating new opportunities in business,
engineering, and many other application areas.
Even though AI & ML technology is integrated into the fabric
of everyday life. The ultimate promises of AI & ML are still
decades away and the necessary advances in knowledge
and technology will require a sustained fundamental
research effort.
32
Download