Sunteți pe pagina 1din 12

So far

Lecture 1: Review of Classical & Modern Control


Lecture 2: MATLAB lecture

In classical & modern control, system are rigidly
modelled, and the software does what is
told. Intelligence comes from the designer
In intelligent control, systems are not rigidly
modelled and the intelligence comes from
the software. Software does what it wants to.



Read the paper
Intelligent control is the applications of :

1. Artificial neural networks
2. Fuzzy Logic
3. Genetic Algorithms
Classifying human breast tissues into normal
and malignant (cancerous)
Pathologically characterized 56 tissue samples were obtained
immediately after surgery.
The tissue samples were irradiated with polarized light of
wavelength 488 nm from a 5W Ar ion laser. Polarized
fluorescence light was collected by using an analyzer. The
fluorescence spectra were recorded using a SPEX 1877
triplemate and a PMT
Data set- 56x190 (190 fluorescence intensities)
28 patients whose normal and malignant breast tissues.


Define artificial neural network yourself after understanding
the example
We have 56 samples. We divide these into training and testing sets.
First take the training samples. We know out of these 28 samples
how many are normal. We will do some computation so that when
we present a normal sample to this black box it will produce a 0
and when we present a malignant sample it will produce a 1 We
use all these 28 training samples to teach the black box. Now we
take a sample from the testing set. When we input a normal sample,
the output must be 0
Black box
(does computation)
190
dimensional
input
One output
which varies
between 0 and 1
Ar
+
LASER
SPEX 1877E
TRIPLEMATE
PMT
PC
POL
DEPOL LENS
ANALYZER
MR
SAMPLE
Experimental set-up for the measurement of
polarized fluorescence spectra
Classification vs. Regression
Classification:
predicts categorical class labels (discrete)
classifies data (constructs a model) based on the
training set and the values (class labels) in a
classifying attribute and uses it in classifying new
data
Regression:
models continuous-valued functions

Classification: identify the class. Basically identify a
discrete quantity
Regression: estimate continuous quantity (tomorrows
temperature, stock market index)
Examples of problems
Input
age, weight, blood
pressure
age, weight, blood
pressure
number of bedrooms, days
on the market , median
price of houses sold in the
area last year
Output
Time on a 5-mile run
(regression)
Finish or not finish a 5-mile
run (Classification)

Selling price of the house
(regression)

Classification: find function that maps inputs into
classes: class = f(x)
Regression: find function that maps inputs into
values: value = f(x)
Inputs are also called features

Features are properties of items to be classified that will aid
in discriminating between classes.


Discrete features: gender, nationality, blood type
Continuous features: location in room, barometric pressure,
value of dollar

If you choose good features, a problem is easy and almost
any solution method works. If you choose badly, nothing
works





Strengths of a Neural Network
Power: Model complex functions, nonlinearity built
into the network
Ease of use:
Learn by example
Very little user domain-specific expertise needed
Intuitively appealing: based on model of biology, will
it lead to genuinely intelligent computers/robots?

Neural networks cannot do anything that cannot be
done using traditional computing techniques, BUT
they can do some things which would otherwise be
very difficult.
1
1 2
2
0
x
1
x
2
Sl.
No
x
1

x
2
y
1
2
3
4
5
6

7
8
0.7
0.8
0.8
1.2
0.6
1.3

0.9
0.9
0.7
0.9
0.25
0.8
0.4
0.5

0.5
1.1
0
1
0
1
0
1

0
1

x
1
x
2
y
1
1
=1.5
x
1
x
2
y
1
1
-1.5
1
0 5 . 1
0 5 . 1 1 1
2 1
2 1
= +
= +
x x
x x
| | 0 5 . 1 1 1
2
1
=
(

x
x
t
0 = +b x w
t
u = =
(

= b b w 5 . 1
1
1
0
u
An Example:
7
8
w
linearly separable
x
1
x
2
Decision Boundary
x
1
x
2
y
Neural network mysticism:
A neural network may solve a practical problem, but it can be
difficult to understand how it solved it. For many problems the
hidden layer is not doing an obvious analysis. If you dont know
what was done, it can be hard to improve it.


Any function can be approximated to arbitrary accuracy by a
network with two hidden layers
Books for Neural Networks
Neural Networks for pattern recognition- C. M. Bishop
Neural Network Design by Martin T. Hagan & Howard B. Demuth & Mark H.
Beale
Neural Networks- A comprehensive foundation- SIMON HAYKIN
Elements of Artificial Neural Networks- K. Mehrotra, C. K. Mohan, Ranka
Introduction to Artificial Neural Systems- J.M. Zurada
Fundamentals of Artificial Neural Networks- Mohamad H Hasssoun
Fundamentals of Neural Networks Architecture, algorithms and applications-
Laurene Fausett
Neural Networks- Algorithms, Applications, and Programming Techniques- James A.
Freeman/ David M. Skapura
An Introduction to Neural Networks- James A. Anderson
Neural Networks- A classroom approach- Satish Kumar (TMH)

S-ar putea să vă placă și