Chapter 1: Supplementary ANS Overview Machine Learning Overview - PowerPoint PPT Presentation

1 / 29

About This Presentation

Title:

Chapter 1: Supplementary ANS Overview Machine Learning Overview

Description:

The network must discover for itself patterns, features, regularities, ... Acquiring a function, based on past inputs and values, from new inputs to values. ... – PowerPoint PPT presentation

Number of Views:20

Avg rating:3.0/5.0

Slides: 30

Provided by: MirellaM3

Category:

more less

Transcript and Presenter's Notes

Title: Chapter 1: Supplementary ANS Overview Machine Learning Overview

1
Chapter 1 Supplementary ANS Overview
Machine Learning Overview

C.S.Chen
2007.03.08

2
Appendix

Machine Learning A Case Study
ANN Overview
Machine Learning Overview

3
Machine Learning A Case Study

Malfunctioning gearboxes have been the cause for
CH-46 US Navy helicopters to crash.
Although gearbox malfunctions can be diagnosed by
a mechanic prior to a helicopters take off, what
if a malfunction occurs while in-flight, when it
is impossible for a human to detect?
Machine Learning was shown to be useful in this
domain and thus to have the potential of saving
human lives!

4
How did it Work?

Consider the following common situation
You are in your car, speeding away, when you
suddenly hear a funny noise.
To prevent an accident, you slow down, and either
stop the car or bring it to the nearest garage.
The in-flight helicopter gearbox fault monitoring
system was designed following the same idea. The
difference, however, is that many gearbox
malfunction cannot be heard by humans and must be
monitored by a machine.

5
So, Wheres the Learning?

Imagine that, instead of driving your good old
battered car, you were asked to drive this truck
Would you know a funny noise from a normal
one?
Well, probably not, since youve never driven a
truck before!
While you drove your car during all these years,
you effectively learned what your car sounds like
and this is why you were able to identify that
funny noise.

6
What did the Computer Learn?

Obviously, a computer cannot hear and can
certainly not distinguish between a normal and an
abnormal sound.
Sounds, however, can be represented as wave
patterns such as this one
which in fact is a series
of real numbers.
And computers can deal with strings of numbers!
For example, a computer can easily be programmed
to distinguish between strings of numbers that
contain a 3 in them and those that dont.

7
What did the Computer Learn? (Contd)

In the helicopter gearbox monitoring problem, the
assumption is that functioning and malfunctioning
gearboxes emit different noises. Thus, the
strings of numbers that represent these noises
have different characteristics.
The exact characteristics of these different
categories, however, are unknown and/or are too
difficult to describe.
Therefore, they cannot be programmed, but rather,
they need to be learned by the computer.
There are many ways in which a computer can learn
how to distinguish between two patterns (e.g.,
decision trees, neural networks, bayesian
networks, etc.) and that is the topics of this
chapter

8
The Real and Artificial Neurons
9

ANNs are systems that are constructed to make
use of some organizational principles resembling
those of the human brains
ANNs are good at tasks such as
pattern matching and classification
function approximation
optimization
vector quantitization
data clustering

10
The Neuron (Processing Element)
11

Models of ANNs are specified by three basic
entitites
Models of the neurons themselves,
Models of synaptic interconnections and
structures,
Training or learning rules for updating the
connecting weights

There are three important parameters about a
neuron
An integrated function associated with the input
of a neuron to calculate the net input (for M-P
neuron)
fi neti
II. A set of links, describing the neuron
inputs, with weights W1, W2, , Wm

13
III. Activation Function

Activation functions a(f) output an activation
value as a function of its net input. Some
commonly used activation functions are
Step Function
Hard limiter (thresold function)
Ramp function
Unipolar sigmoid function
Bipolar sigmoid function

14
Learning Rules

Generally, we can classify learning in ANNs in
two broad classes
Parameter learning which is concerned with
updating of the connecting weights
Structure learning which focuses on the change
in the network structure, including the number of
PEs and their connection types.
These two kinds of learning can be performed
simultaneously or separately.

In weight learning, we have to develop learning
rules to efficiently guide the weight matrix W in
approaching a desired matrix that yields the
desired network performance. In general, learning
rules are classified into three categories
Supervised Learning (learning with a teacher)
Reinforcement Learning (learning with a critic)
Unsupervised learning

16
In supervised learning when input is applied to
an ANN, the corresponding desired response of the
system is given. An ANN is supplied with a
sequence of examples (x1, d1) , (x2, d2)...(xk,
dk) of desired input-output pairs. In
reinforcement learning only less detailed
information than supervised learning is
available. There is only a single bit of feedback
information indicating whether the output is
right or wrong. That is, it just says how good or
how bad a particular output is and provides no
hint as to what the right answer should be.
17

In unsupervised learning,
There is no teacher to provide any feedback
information. The network must discover for itself
patterns, features, regularities, correlations or
categories in the input data and code for them in
the output.
While discovering these features, the network
undergoes changes in its parameters this process
is called self-organizing.

18
Three Categories of Learning
Supervised Learning
Reinforcement Learning
Unsupervised Learning
19
What is Machine Learning?

Machine Learning allows computers to learn from
their experiences and from gathered data
Machines are good at gathering data and
performing complex analysis
The goal of machine learning is to build
computer systems that can adapt and learn from
their experience.
Tom Dietterich

20
Machine Learning

Learning
Acquiring a function, based on past inputs and
values, from new inputs to values.
Learn concepts, classifications, values
Identify regularities in data
Learning as Search
A hypothesis is a guess at a function that can be
used to account for the inputs.
A hypothesis space is the space of all possible
candidate hypotheses.
Learning is a search through the hypothesis space
for a good hypothesis.

21
Another Definition of Machine Learning

Machine Learning algorithms discover the
relationships between the variables of a system (
input, output and hidden ) from direct samples of
the system
These algorithms originate form many fields
Statistics, mathematics, theoretical computer
science, physics, neuroscience, etc

22
A Generic System
Input Variables
Hidden Variables
Output Variables
23
Inductive Learning

Suppose the underlying problem domain is
described by a function f
Given pairs ltx, f(x)gt
Compute a hypothesis h that approximates f as
well as possible given the presented data
In general the input under-constrains the
function h, so we have to choose. The way that
choice is performed is called bias.

24
Inductive learning method