Artificial Neural Network Paper Presentation

www.studentyogi.com www.studentyogi.com
Artificial Neural Network
INTRODUCTION

BACKGROUND:

co om
Many task which seem simple for us, such as reading a
handwritten note or recognizing a face, are difficult task for even the most

m
advanced computer. In an effort to increase the computer ability to perform
such task, programmers began designing software to act more like the
human brain, with its neurons and synaptic connections. Thus the field of

gi. .c
“Artificial neural network” was born. Rather than employ the traditional
method of one central processor (a Pentium) to carry out many instructions
one at a time, the Artificial neural network software analyzes data by passing

oogi
it through several simulated processos which are interconnected with
synaptic like “weight”
Once we have collected several record of the data we wish to
analyze, the network will run through them and “learn ” the input of each
record may be related to the result. After training on a few doesn’t cases the
ntyy
network begin to organize and refines its on own architecture to feed the
eent
data to much the human brain; learn from example.
This “REVERSE ENGINEERING” technology were once regarded
as the best kept secret of large corporate, government an academic
researchers.
The field of neural network was pioneered by BERNARD
t t dd

WIDROW of Stanford University in 1950’s.
ssuu
w. .
w
ww
ww

www.studentyogi.com www.studentyogi.com1

Why would anyone want a `new' sort of computer?

What are (everyday) computer systems good at... .....and not so good at?
Good at Not so good at

om
Interacting with noisy data or data
Fast arithmetic
from the environment
Doing precisely what the programmer
Massive parallelism
programs them to do

i.c
Massive parallelism
Fault tolerance
Adapting to circumstances

og
Where can neural network systems help?

• where we can't formulate an algorithmic solution.
nty
• where we can get lots of examples of the behaviour we require.
• where we need to pick out the structure from existing data.

What is a neural network?
Neural Networks are a different paradigm for computing:
de

• Von Neumann machines are based on the processing/memory
abstraction of human information processing.
stu

• Neural networks are based on the parallel architecture of animal
brains.

Neural networks are a form of multiprocessor computer system, with
w.

• Simple processing elements
• A high degree of interconnection
• Simple scalar messages
• Adaptive interaction between elements
ww

Artificial neural network (ANNs) are programs designed to solve
any problem by trying to mimic structure and function of our nervous
system. Neural network are based on simulated neurons. Which are joined
together in a variety of ways to form networks. Neural network resembles
the human brain in the following two ways: -

A neural network acquires knowledge through learning
A neural network’s knowledge is stored with in the interconnection strengths
known as synaptic weight.
Neural network are typically organized in layers. Layers are

om
made up of a number of interconnected ‘nodes’, which contain an ‘activation
function’. Patterns are presented to the network via the ‘input layer’, which
communicates to one or more ‘hidden layers’ where the actual processing is
done via a system of weighted ‘connections’. The hidden layers then link to
an ‘output layer’ where the answer is output as shown in the graphic below.

i.c
Hidden Layers

og
Connections nty
de

Input layer Output layer

Basic structure of neural network
stu

Each layer of neural makes independent computation on data that it receives
and passes the result to the next layers(s). The next layer may in turn make
independent computation and pass data further or it may end the
w.

computation and give the output of the overall computation .The first layer is
the input layer and the last one, the output layer. The layers that are placed
within these two are the middle or hidden layers.
A neural network is a system that emulates the cognitive
ww

abilities of the brain by establishing recognition of particular inputs and
producing the appropriate output. Neural networks are not “hard-wired” in
particular way; they are trained using presented inputs to establish their own
internal weights and relationships guided by feedback. Neural networks are
free to form their own internal working and adapt on their own.
Commonly neural network are adjusted, or trained so that a particular
input leads to a specific target output

Target

Neural networksddss
Neural network

om
Including
Input connections Compare
(called weights)

i.c
Figure showing adjust of neural network

og
There, the network is adjusted based on a comparison of the output and the
target, until the network output matches the target. Typically many such
input/target pairs are used to train network.
nty
Once a neural network is ‘trained’ to a satisfactory level it may be used as an
analytical tool on other data. To do this, the user no longer specifies any
training runs and instead allows the network to work in forward propagation
mode only. New inputs are presented to the input pattern where they filter
de

into and are processed by the middle layers as though training were taking
place, however, at this point the output is retained and no back propagation
occurs.
The structure of Nervous system
stu

Nervous system of a human brain consists of neurons, which are
interconnected to each other in a rather complex way. Each neuron can be
thought of as a node and interconnection between them are edge, which has
a weight associates with it, which represents how mach the tow neuron
w.

which are connected by it can it interact.

Node (neuron)
ww

Edge
(interconnection)


Functioning of A Nervous System
The natures of interconnections between 2 neurons can such that – one
neuron can either stimulate or inhibit the other. An interaction can take place
only if there is an edge between 2 neurons. If neuron A is connected to

om
neuron B as below with a weight w, then if A is stimulated sufficiently, it
sends a signal to B. The signal depends on

w

i.c
A B
The weight w, and the nature of the signal, whether it is stimulating or
inhibiting. This depends on whether w is positive or negative. If its
stimulation is more than its threshold. Also if it sends a signal, it will send it

og
to all nodes to which it is connected. The threshold for different neurons
may be different.
If many neurons send signal to A, the combined stimulus may be more than
the threshold.
nty
Next if B is stimulated sufficiently, it may trigger a signal
to all neurons to which it is connected.
Depending on the complexity of the structure, the
overall functioning may be very complex but the functioning of individual
de

neurons is as simple as this. Because of this we may dare to try to simulate
this using software or even special purpose hardware.

Major components Of Artificial Neuron
stu

This section describes the seven major components, which
make up an artificial neuron. These components are valid whether the
neuron is used for input, output, or is in one of the hidden layers.
Component 1. Weighing factors:
w.

A neuron usually receives many simultaneous inputs. Each input
has its own relative weight, which gives the input the impact that it needs on
the processing elements summation function. These weights perform the
same type of function, as do the varying synaptic strengths of biological
ww

neurons. In bath cases, some input are made more important than others so
that they have a greater effect on the processing element as they combine to
produce a neuron response.
Weights are adaptive coefficients within the network that determine
the intensity of the input signal as registered by the artificial neuron. They
are a measure of an input’s connection strength. These strengths can be


modified in response to various training sets and according to a network’s
specific topology or through its learning rules.
Component 2. Summation Function:
The first step in a processing element’s operation is to compute the

om
weighted sum of all of the inputs. Mathematically, the inputs and the
corresponding weights are vectors which can be represented as (i1,
i2,………….in) and (w1, w2,……………….wn). The total input signal is
the dot, or inner, product of these two vectors. This simplistic summation
function is found by multiplying each component of the i vector by the

i.c
corresponding component of the w vector and then adding up all the
products. Input1 = i1*w1, input2=i2*w2, etc., are added as
input1+input2+………..+inputn. The result is a single number, not a multi-
element vector.

og
Geometrically, the inner product of two vectors can be considered a
measure of their similarity. If the vector point in the same direction, the
inner product is maximum; if the vectors points in opposite direction (180
nty
degrees out of phase), their inner product is minimum.
The summation function can be more complex than just the simple input
and weight sum of products. The input and weighting coefficients can be
combined in many different ways before passing on to the transfer function.
In addition to a simple product summing, the summation function can select
de

the minimum, maximum, majority, product, or several normalizing
algorithms. The specific algorithm for combining neural inputs is
determined by the chosen network architecture and paradigm.
Component 3: Transfer Function:
stu

The result of the summation function, almost always the weighted sum,
is transformed to a working out put through an algorithm process known as
the transfer function. In transfer function the summation total can be
compared with some threshold to determine the neural output. If the sum is
w.

greater than the threshold value, the processing element generates a signal. If
the sum of the input and weight product is less than the threshold, no signal
(no some inhibitory signal) is generated. Both types of response are
significant.
ww

A simple binary neuron with 3 inputs ( X [1], X [2] and X [3]) and 2
outputs (O [1] and O[2] ). Every neuron has a particular threshold. A neuron
fires only if the weighted sum of its input exceeds the threshold.
Sum=summation (W i * Xi )
If sum>=T then the output is
Output O[1] and O[2]= 1 if ( ∑ X[i] * W[i] >=T )


=0 else
X [1]
W [1]

om
X [2] W [2] (synaptic connection)
O [1]

X [3] W [3]
(neuron processing node)

i.c
McCulloch-Pitts Neuron Model

og
We can use different kind of threshold functions namely:
Sigmoid function : f (x)=1/(1+exp(-x))
Step function : f (x) =0 if x<T
=k if x>=T
nty
Ramp funtion : f (x)=ax+b

The transfer function could be something as simple as depending upon
whether the result of the summation function is positive or negative. The
de

network could output zero and one, and minus one, or other numeric
combinations. The transfer function would then be a “hard limiter” or step
function.
stu

Hard limiter Ramping function
Y y 1
1 x
x
-1 -1
w.

x<0, y=-1 x<0, y=0
x>=0, y=1 0<=x<=1, y=x
x>1, y=1
ww

y sigmoid functions y
1 1
x x
-1,0
y = 1/(1+e-x) x>=0, y=1-1(1+x)
x<0, y=-1+1(1-x)

Figure : Sample Transfer Functions


Component 4: Scaling and Limiting:
After the processing element’s transfer function, the result can pass
through additional processes which scale and limit. This scaling simply

om
multiplies a scale factor times the transfer value, and then adds an offset.
Limiting is the mechanism, which ensures that the scaled result does not
exceed, and upper or lower bound. This limiting is in addition to the hard
limits that the original transfer function may have performed.
Component 5: Output function (competition)

i.c
At Each processing element is allowed one output signal, which it
may output to hundreds of other neurons. This is the just like the biological
neuron, where there are many inputs and only one output action. Normally,
the output is directly equivalent to the transfer function’s result. Some

og
network topologies, however, modify the transfer result to incorporate
competition among neighboring processing elements. Neurons are allowed
to complete with each other, inhibiting processing elements unless they have
great strength. Competition can occur one or both of two levels. First,
nty
competition determines which artificial neuron will be active, are provides
an output. Second, competitive inputs help determine which processing
elements will participate in the learning or adaptation process.
Component 6: Error function and back-propagated value:
de

In most learning networks the difference between the current output
and the desired output is calculated. This raw error is then transformed by
the error function to match particular network architecture. The most basic
architectures use this error directly, but some square the error while retaining
stu

its sign, some cube the error, other paradigms modify the raw error to fit
their specific purposes. The artificial neuron’s error is then typically
propagated into the learning function of another processing element. This
error term is sometimes called the current error.
w.

The current error is typically propagated backwards to a previous layer.
Yet, this back-propagated value can be either the current error, the current
error scaled in some manner (obtained by the derivative of the transfer
function), or some desired output depending on the network type. Normally,
ww

this back-propagated value, after being scaled by the learning function, is
multiplied against each of the incoming connection weights to modify them
before the next learning cycle.
Component 7: Learning function :
The purpose of the learning function is to modify the variable
connection weights on the inputs of each processing elements according to
some neural base algorithm. This process of changing the weights of the

inputs connections to achieve some desired results could also be called the
adaption function, as well as the learning mode.

Paradigms of learning—

om
There are three broad paradigms of learning:
• Supervised
• Unsupervised (or self-organised)
• Reinforcement learning ( a special case of supervised learning ).

i.c
Supervised learning:
The vast majority of the artificial neural network solutions have
been trained with supervision. In this mode, the actual output of a neural

og
network is compared to the desired output. The network then adjusts
weights, which are usually randomly set to begin with, so that the next
iteration, or cycle, will produce a closer match between the desired and
the actual output. This learning method tries to minimize the current
nty
errors of all processing elements. This global error reduction is created
over time by continuously modifying the input weights until acceptable
network accuracy is reached.

Adaptive
de

network O
X W
stu

Distant Learning signal
generater D

Ρ [d, 0]
Distance measure
w.

Block dig. Of supervised learning
ww

Unsupervised learning:
In supervised learning the system directly compares the network
output with a known correct or desired answer, whereas in unsupervised
learning the output is not known. Unsupervised training allows the
neurons to compete with each other until winner emerge. The resulting
values of the winner neurons determine the class to which a particular


data set belongs. Unsupervised learning is the great promise of the future.
It shouts that computers could some day learn on their own in a true
robotic sense. Currently, this learning method is limited to networks
known as self organizing map. These kinds of networks are not in wide

om
spread use.

Adaptive
network
x W o

i.c
og
Block diagram of unsupervised learning
nty
Reinforcement learning:
Reinforcement learning is a form of supervised learning where
adopted neuron receives feedback from the environment that directly
influences learning.
de

Learning law :
The following general learning rules is adopted in the neural network
studies:
stu

The weight vector wi =[Wi1, Wi2……………………………….Win] t
increases proportion to the product of input x and learning signal r. The
learning signal r is a function of wi x, and sometimes of the teacher’s signal
di.
Hence we have, r =r (Wi, X, di) and increment in weight vector produced
w.

by the learning step at time t is ∆Wi (t) =cr [Wi (t), X (t), di (t)] X (t)
Where c is learning constant.
Thus Wi (t+1) =Wi (t)+ cr [Wi (t), X (t), di (t)] X (t)
ww

Various learning rules are their assist the learning process.
They are :
1. Hebbian learning rule :
This rule represents purely feed forward, unsupervised learning

10

th
X1 Wi1 i neuron

X2 Wi2 Oi

om
Wi3 ∆Wi
X3 Learning di
Win x signal
r generater
Xn

i.c
c
Hebbian Learning Rule

According to this rule, we have

og
r=f(Wti, X)
and increment of weight becomes
∆Wi=c f(Wti, X) X
nty
In this learning, the initial weight of neuron is.
2. Perceptron learning rule:
This learning is supervised and learning signal is equal to
r =di-Oi
Where Oi=sgn(Wti, X) and di is the desired response.
de

Weight adjustment in this method is
∆Wi=c[di- sgn(Wti, X)] X
stu

in this method of learning, initial weight can have any value and neuron
much be binary bipolar are binary unipolar.

X1
X2 Wi1 TLU
w.

Wi2 neti
+ oi
ww

x3 Wi3 ∆Wi

Win di-oi + di
Xn X

C
Perception Learning Rule
11

3. Delta learning rule:
This rule is valid for continuous activation functions and in the
supervised training mode. The learning signal is called as delta and is
given as:

om
r=[di – f(Wti X)] f ′ (Wti X)
The adjustment for the single weight in this rule is given as:
∆Wi = c (di-Oi) f ′ (net i) X
In this method of learning, the initial weight can have any value and the
neuron must be continuous.

i.c
X1
Wi1 continuous perception

og
X2 wi2 f(neti) Oi
Win
Xn ∆Wi f΄(net i)
+
nty
X
r di - oi + di

c
Delta Learning Rule
de

4. Windrow-Hoff learning rule:
This is applicable for the supervised training of neural networks and is
stu

independent of activation function.
Learning signal is given as :
r = di – Wti X
The weight vector increment under this learning rule is :
∆Wi=c (di - Wti X) X
w.

5. Correlation learning rule :
Substituting r = di in general learning rule, we obtain correlation learning
rule. The adjustment for the weight vector is given by:
ww

∆Wi = c di X
6. Winner-take all learning rule:
This rule is applicable for an ensemble of neurons, let’s say being
arranged in a layer of p units. This learning is base on the premise that
one of the neurons in the layer, say the mth, has the maximum response
due to input x, as shown in fig. This neuron is declared the winner. As a
result of this winning event, the weight vector Wm containing weights
12

W11
X1 Wm1 O1
Wp1 “winning neuron”
W1j
Xj Wmj Om

om
Wpj

W1n Wmn
Xn Wpn Op

i.c
Winner –take –All learning Rule
highlight in the fig. , by double headed arrow, is the only one adjusted in
the given unsupervised learning. It’s increment is computed as follows:
∆Wm = α (X - Wm)

og
7.outstar learning rule:
This is an another learning rule that is best explained when the neurons
are arranged in layers. This rule is designed to produce a desired response
nty
d of the layer of p neurons as shown in fig… This rule is concerned with
the supervised learning and the weight adjustment is computed as:
∆Wj = β (X - Wj)
de

X W11 o1
Wm1
stu

Wp1 ∆Wi +
+ d1
W1j β
Xj Wmj om
Wpj
w.

∆Wi +
+ dm
W1n Wmn β
ww

Xn op
Wpn ∆Wi +
+ dp
β

Outstar Learning Rule

13

Training a neural network:
Since the output of the neural network may not be what is
expected, the network needs to be trained. Training involves altering the
interconnection weights between the neurons. A criterion is needed to

om
specify when to change the weights and how to change them. Training is
an external process whiling learning is the process that takes place
internal to the network. The following guideline will be of help as a step
methodology for training a network.
• Choosing the number of neurons

i.c
The number of hidden neurons affect how well the network is able
to separate the data. A large number of hidden neurons will ensure
correct learning and the network is able to correctly predict the data it has

og
been trained on, but its performance on new data, its ability to generalize,
is compromised. With too few hidden neurons, the network may be
unable to learn the relationship amongst the data and the error will fail to
fall below an acceptable level. Thus, selection of the number of hidden
nty
neurons is a crucial decision. Often a trial and error approach is taken
starting with a modest number of hidden neurons and gradually
increasing the number if the network fails to reduce its error. A much
used approximation for the number of hidden neurons for a three layered
network is N=1/2(j + k)+v P, where J and K are the number of input
de

neurons and P is the number of patterns in the training set.
• Choosing the initial weights
The learning algorithm uses steepest descent technique, which rolls
stu

straight downhill in weight space until the first valley is reached. This
valley may not correspond to a zero error for the resulting network. This
makes the choice of initial starting point in the multidimensional weight
space critical. However, there are no recommended rules for this
selection except trying several different weight values to see if the
w.

network results are improved.
• Choosing the learning rate
Learning rate effectively controls the size of the step that is taken
ww

in multidimensional weight space when each weight is modified. If the
selected learning rate is too large then the local minimum may be over
stopped constantly, resulting in oscillations and slow convergence to lower
error state. If the learning rate is too low, the number of iterations requires
may be too large, resulting in slow performance. Usually the default value of
most commercial neural network packages are in the range 0.1-0.3 providing

14

a satisfactory balance between the two reducing the learning rate may help
improve convergence to a local minimum of the error surface.
• Choosing the activation function
The learning signal is a function of the error multiplied by the

om
gradient of the activation function df/d (net). The larger the value
of the gradient, the more weight the learning will receive. For
training, the activation function must be monotonically increasing
from left to right, differentiable and smooth.

i.c
Models Of artificial Neural Networks
There are different kinds of neural network models that can be used.
Some of the common ones are:
1. Perception model:

og
This is a very simple model and consists of a single ‘trainable’
neuron. Trainable means that its threshold and input each input has a
desired output (determined by us). If the neuron doesn’t gives the desired
nty
output, then it has made a mistake. To rectify this, its threshold and/or
input weights must be changed. How this change is to be calculated is
determined by the learning algorithm.
The output of the perceptron is constrained to Boolean values – (true,
false), (1, 0), (1, -1) or whatever. This is not a limitation because if the
de

output of the perceptron were to be the input for something else, then the
output edge could be made to have a weight. Then the output would be
dependant on this weight.
The perceptron looks like –
stu

X1
W1
X2 W2
w.

X3 W2
y

Xn Wn
ww

X1, X2, …………., Xn are inputs. These could be real numbers or Boolean
values depending on the problem.
y is the output and is Boolean.
w 1, w2, …………, wn are weights of the edges and are real valued.
T is the threshold and is a real valued.
The output y is 1 if the net input which is :
w1 x1 + w2 x2 + …….+ wn xn
15

is greater than the threshold T. Otherwise the output is zero.
2. Feed – Forward Model:
Elementary feed forward architecture of m neurons and
receiving n inputs is shown in the figure. Its output and input vectors are

om
respectfully.
O = [O] O2 ………Om]
X = [x] x2………..xn]
Weight Wij connects the ith neuron with the jth input. Hence activation
value net i for the ith neuron is

i.c
Net i = j=1∑n Wij Xj for i=1, 2, 3,……, n
Hence, the output is given as
Oi = f(Wti X) for i = 1, 2, 3, ……….., m
Where Wi is the weight vector containing the weights leading towards

og
the ith output node and defined as
Wi = [Wi1 Wi2 …………. Win]
If  is the nonlinear matrix operarte, the mapping of input space X to
nty
output space O implemented by the network can be expressed
O = WX
Where W is the weight matrix, also called the connection matrix.
The generic feedforward network is characterized by the lack of
feedback. This type network can be connected in cascade to create a
de

multiplayer network.
W11
X1 W21 O1
W12 1
stu

X2 W22
w.

W2n O2
Wm1 2
W1n Wm2
ww

Xn Wmn m On

X(t) ┌ [Wx] O(t)

Single –layer Feed forward network:interconnection scheme & block dig.
16


W11 W12 ……..………W1n
W21 W22 ……..………W2n

om
. . ……………. .
W= . . …………… .
. . …………… .

Wm1 …………….Wmn

i.c
f(.) O…………..O
O f (.) ………..O

og
┌ [.] =
. . ………… .
. . ………… .
nty
O O………….f(.)

3. Feed-Back model:
A feedback network can be obtained from the feedforawrd
de

network by connecting the neuron’s outputs to their inputs, as shown in
the fig. The essence of closing the feedback loop is to hold control of
output Oi through outputs Oj; for j =1, 2, ……, m. or controlling the
output at instant t+∆ by the output at instant t. This delay ∆ is
stu

introduced by the delay element in the feedback loop. The mapping of
O(t) into O(t+ ∆) can now be written as
O(t+ ∆) = ┌[W o(t)]
w.
ww

17

O2 (t)
∆
O1 (t)
W11 ∆

om
X1 W21 O1(t+ ∆)
W12

x2 W22 O2(t+ ∆)

i.c
W2n
Lag free neuron
W1n Wm1
Wm2

og
Xn Wmn O3(t+ ∆)
m
∆
Om(t)
nty
Instantaneous
X (0) network O (t+∆)
┌ [W o(t)]
de

O (t)
stu

Delay
∆

Single –layer feedback network : interconnection scheme & block dig.
w.

Notations Used:
M1 is a 2-D matrix where M1[i] [j] represents the weight on the
connection from the ith input neuron to the jth neuron in the hidden layer.
ww

M2 is a 2 –D matrix where M2[i][j] denotes the weight on the
connection from the ith hidden layer neuron to the jth output layer neuron.
x, y and z are used to denote the outputs of neurons from the input
layer, hidden layer and output layer respectively.
If there are m input data, then (x1, x2, ……., xm). P denotes the desired
output pattern with components (p1, p2, ………, pr) for r outputs.
Let the number of hidden layer neurons be n.
18

βo = learning parameter for the output layer.
βh = learning parameter for the hidden layer.
α = momentum term
θj = threshold value (bias) for the jth hidden layer neuron.
τj = threshold value for the jth output layer neuron.

om
ej = error at the jth output layer neuron.
tj = error at the jth hidden layer neuron.
Threshold function = sigmoid function : F(x) = 1/(1 + exp(x)).
Mathematical expressions:

i.c
Output of jth hidden layer neuron : yj = f((Σi xi M1[i][j]) + θj)
Output of jth output layer neuron : zj = f((Σi yi M2[i][j]) + τj).
ith component of vector of output differences:
desired value – computed value = Pj – zj

og
th
i component of output error at the output layer:
ej = pj – zj.
th
i component of output error at the hidden layer:
ti = yi (1 – yi ) (Σj M2[i][j] ej )
nty
Adjustment of weights between the ith neuron in the hidden layer and jth
output neuron:
∆M2[i][j] (t) = β0 yi ej + α ∆M2 [i][j] (t - 1)
Adjustment of weights between the ith input neuron and jth neuron in the
de

hidden layer :
∆M1[i][j] = βh xi tj + α ∆M1 [i][j] (t - 1)
Adjustment of the threshold value for the jth output neuron:
∆ τj = β0 ej
stu

Adjustment of the threshold value for the jth hidden layer neuron:
∆ θj = βh ej

Neural network applications:
w.

Aerospace
• High performance aircraft autopilot, flight path simulation, aircraft
ww

control systems, autopilot enhancements, aircraft component
simulation, aircraft component fault detection.
Automobile control
• Automobile automatic guidance system, warranty activity analysis.
Banking

19

• Check and other document reading credit application
evaluation.
Credit card activity checking
• Neural networks are used to spot unusual credit card

om
activity that might possibly be associated with loss of a
credit card
Defense
• Weapon steering, target tracking, object discrimination,

i.c
facial recognition, new kinds of sensors, sonar, radar and
image signal processing including data compression,
feature extraction and noise suppression, signal/ image

og
identification.
Electronics
• Code sequence prediction, integrated circuit chip laying,
nty
process control, chip failure analysis, machine vision
voice synthesis, nonlinear modeling.
Entertainment
• Animation, special effects, market forecasting.
de

Financial
• Real estate appraisals, loan advisor, mortgage screening,
corporate bond rating, credit-line use analysis, and
stu

portfolio trading program, corporate financial analysis,
and currency price prediction.
Industrial
• Neural networks are being trained to predict the output
w.

gasses of furnaces and other industrial process. They then
replace complex and costly equipment used for this
purpose in the past.
ww

Insurance
• Neural networks are used in policy application evaluation,
product optimization.
Manufacturing
• Neural networks are used in manufacturing process
control, product design and analysis, process and machine
20

diagnosis, real-time particle identification, visual quality
analysis, paper quality prediction, computer – chip quality
analysis, analysis of grinding operations, chemical product
design analysis, machine maintenance analysis, project

om
bidding, planning and management, dynamic modeling of
chemical process system.
Medical
• Neural networks are used in breast cancer cell analysis,

i.c
EEG and ECG analysis, prosthesis design, optimization of
transplant times, hospital expense reduction, hospital
quality improvement, and emergency room test

og
advisement.
Oil and Gas
• Neural networks are used in exploration of oil and gas.
nty
Robotics
• Neural networks are used in trajectory control, forklift
robot, manipulator controllers, vision systems
Other application
• Artificial intelligence
de

• Character recognition
• Image understanding
• Logistics
stu

• Optimization
• Quality Control
• Visualization
w.

Advantages and disadvantages Of ANN:
Advantage: --
ww

1. It involve human like thinking.
2. They handle noisy or missing data.
3. They create their own relationship amongst information – no
equations!
4. They can work with large number of variables or parameters.
5. They provide general solutions with good predictive accuracy.
6. System has got property of continuous learning.
21

7. They deal with the non-linearity in the world in which we live.
Disadvantage –
1. In learning process time required may be in several months or years.
2. It is hard to implement.

om
Conclusion –
In this present world of automation to automate systems neural
network and fuzzy logic systems are required. Fuzzy logic deals with
vagueness or uncertainty and neural network related to human like thinking.

i.c
If we use only fuzzy systems or only NN complete automation is never
possible. The combination is suitable because fuzzy logic has tolerance for
impression of data, while neural networks have tolerance for noisy data. As
there are different advantages and disadvantages usage of NN thus we use

og
combination of neural networks and Fuzzy logic in order to implement a real
world system without manual interference.
Artificial neural network, in the present scenario is novel in it’s
technological field and still we have to witness a large of it’s development in
nty
the upcoming era’s, whose speculations are not required, as it will speak for
themselves.
de

Bibliography
stu

References books:--
• “Introduction to Artificial Neural Network”, by Jacek M. Zurada;
Jaico Publishing House, 1999.
• “An Introduction to Neural Network”, by James A. Anderson; PHI,
w.

1999.
• “Elements of Artificial Neural Network”, K. Mehrotra, C.K. Mohan
and Sanjay ranka, MIT Press, 1997.
• “Neural Network nad Fuzzy System”, by Bart Kosko; PHI, 1992.
ww

• “Neural Network – A comprehensive foundation”, Simon Haykin,
Macmillan Publishing Co., Newyork, 1993.
Reference sites –
• http://www.cs.stir.ac.uk/~lss/NNInro/invSlides.htm
• http://www.bitstar.com/nnet.htm

22

• http://www.pmsi.fr/sxcxmpa.htm
• http://www.pmsi.fr/neurinia.htm

om
i.c
og
nty
de
stu
w.
ww

23

Artificial Neural Network Paper Presentation

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Artificial Neural Network Paper Presentation (20)

More from guestac67362 (20)

Recently uploaded (20)

Artificial Neural Network Paper Presentation