Fundamentals of Neural Network (Soft Computing)

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
Fundamentals of Neural Networks : Soft Computing Course Lecture 7 – 14, notes, slides
www.myreaders.info/ , RC Chakraborty, e-mail rcchak@gmail.com , Dec. 01, 2010
http://www.myreaders.info/html/soft_computing.html
Fundamentals of Neural Networks
Soft Computing
www.myreaders.info
Return to Website
Neural network, topics : Introduction, biological neuron model,
artificial neuron model, neuron equation. Artificial neuron : basic
elements, activation and threshold function, piecewise linear and
sigmoidal function. Neural network architectures : single layer feed-
forward network, multi layer feed-forward network, recurrent
networks. Learning methods in neural networks : unsupervised
Learning - Hebbian learning, competitive learning; Supervised
learning - stochastic learning, gradient descent learning; Reinforced
learning. Taxonomy of neural network systems : popular neural
network systems, classification of neural network systems as per
learning methods and architecture. Single-layer NN system : single
layer perceptron, learning algorithm for training perceptron,
linearly separable task, XOR problem, ADAptive LINear Element
(ADALINE) - architecture, and training. Applications of neural
networks: clustering, classification, pattern recognition, function
approximation, prediction systems.

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
Soft Computing
Topics
(Lectures 07, 08, 09, 10, 11, 12, 13, 14 8 hours)
Slides
1. Introduction
Why neural network ?, Research History, Biological Neuron model,
Artificial Neuron model, Notations, Neuron equation.
03-12
2. Model of Artificial Neuron
Artificial neuron - basic elements, Activation functions – Threshold
function, Piecewise linear function, Sigmoidal function, Example.
13-19
3. Neural Network Architectures
Single layer Feed-forward network, Multi layer Feed-forward network,
Recurrent networks.
20-23
4. Learning Methods in Neural Networks
Learning algorithms: Unsupervised Learning - Hebbian Learning,
Competitive learning; Supervised Learning : Stochastic learning,
Gradient descent learning; Reinforced Learning;
24-29
5. Taxonomy Of Neural Network Systems
Popular neural network systems; Classification of neural network
systems with respect to learning methods and architecture types.
30-32
6. Single-Layer NN System
Single layer perceptron : Learning algorithm for training Perceptron,
Linearly separable task, XOR Problem; ADAptive LINear Element
(ADALINE) : Architecture, Training.
32-39
7. Applications of Neural Networks
Clustering, Classification / pattern recognition, Function approximation,
Prediction systems.
39
8. References : 40
02

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
What is Neural Net ?
• A neural net is an artificial representation of the human brain that
tries to simulate its learning process. An artificial neural network
(ANN) is often called a "Neural Network" or simply Neural Net (NN).
• Traditionally, the word neural network is referred to a network of
biological neurons in the nervous system that process and transmit
information.
• Artificial neural network is an interconnected group of artificial neurons
that uses a mathematical model or computational model for information
processing based on a connectionist approach to computation.
• The artificial neural networks are made of interconnecting artificial
neurons which may share some properties of biological neural networks.
• Artificial Neural network is a network of simple processing elements
(neurons) which can exhibit complex global behavior, determined by the
connections between the processing elements and element parameters.
03

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network – Introduction
1. Introduction
Neural Computers mimic certain processing capabilities of the human brain.
- Neural Computing is an information processing paradigm, inspired by
biological system, composed of a large number of highly interconnected
processing elements (neurons) working in unison to solve specific problems.
- Artificial Neural Networks (ANNs), like people, learn by example.
- An ANN is configured for a specific application, such as pattern recognition or
data classification, through a learning process.
- Learning in biological systems involves adjustments to the synaptic
connections that exist between the neurons. This is true of ANNs as well.
04

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
1.1 Why Neural Network
Neural Networks follow a different paradigm for computing.
■ The conventional computers are good for - fast arithmetic and does
what programmer programs, ask them to do.
■ The conventional computers are not so good for - interacting with
noisy data or data from the environment, massive parallelism, fault
tolerance, and adapting to circumstances.
■ The neural network systems help where we can not formulate an
algorithmic solution or where we can get lots of examples of the
behavior we require.
■ Neural Networks follow different paradigm for computing.
The von Neumann machines are based on the processing/memory
abstraction of human information processing.
The neural networks are based on the parallel architecture of
biological brains.
■ Neural networks are a form of multiprocessor computer system, with
- simple processing elements ,
- a high degree of interconnection,
- simple scalar messages, and
- adaptive interaction between elements.
05

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
1.2 Research History
The history is relevant because for nearly two decades the future of
Neural network remained uncertain.
McCulloch and Pitts (1943) are generally recognized as the designers of the
first neural network. They combined many simple processing units together
that could lead to an overall increase in computational power. They
suggested many ideas like : a neuron has a threshold level and once that
level is reached the neuron fires. It is still the fundamental way in which
ANNs operate. The McCulloch and Pitts's network had a fixed set of weights.
Hebb (1949) developed the first learning rule, that is if two neurons are
active at the same time then the strength between them should be
increased.
In the 1950 and 60's, many researchers (Block, Minsky, Papert, and
Rosenblatt worked on perceptron. The neural network model could be
proved to converge to the correct weights, that will solve the problem. The
weight adjustment (learning algorithm) used in the perceptron was found
more powerful than the learning rules used by Hebb. The perceptron caused
great excitement. It was thought to produce programs that could think.
Minsky & Papert (1969) showed that perceptron could not learn those
functions which are not linearly separable.
The neural networks research declined throughout the 1970 and until mid
80's because the perceptron could not learn certain important functions.
Neural network regained importance in 1985-86. The researchers, Parker
and LeCun discovered a learning algorithm for multi-layer networks called
back propagation that could solve problems that were not linearly
separable.
06

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
1.3 Biological Neuron Model
The human brain consists of a large number, more than a billion of
neural cells that process information. Each cell works like a simple
processor. The massive interaction between all cells and their parallel
processing only makes the brain's abilities possible.
Fig. Structure of Neuron
Dendrites are branching fibers that
extend from the cell body or soma.
Soma or cell body of a neuron contains
the nucleus and other structures, support
chemical processing and production of
neurotransmitters.
Axon is a singular fiber carries
information away from the soma to the
synaptic sites of other neurons (dendrites
and somas), muscles, or glands.
Axon hillock is the site of summation
for incoming information. At any
moment, the collective influence of all
neurons that conduct impulses to a given
neuron will determine whether or not an
action potential will be initiated at the
axon hillock and propagated along the axon.
Myelin Sheath consists of fat-containing cells that insulate the axon from electrical
activity. This insulation acts to increase the rate of transmission of signals. A gap
exists between each myelin sheath cell along the axon. Since fat inhibits the
propagation of electricity, the signals jump from one gap to the next.
Nodes of Ranvier are the gaps (about 1 µm) between myelin sheath cells long axons
are Since fat serves as a good insulator, the myelin sheaths speed the rate of
transmission of an electrical impulse along the axon.
Synapse is the point of connection between two neurons or a neuron and a muscle or
a gland. Electrochemical communication between neurons takes place at these
junctions.
Terminal Buttons of a neuron are the small knobs at the end of an axon that release
chemicals called neurotransmitters.
07

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Information flow in a Neural Cell
The input /output and the propagation of information are shown below.
Fig. Structure of a neural cell in the human brain
■ Dendrites receive activation from other neurons.
■ Soma processes the incoming activations and converts them into
output activations.
■ Axons act as transmission lines to send activation to other neurons.
■ Synapses the junctions allow signal transmission between the
axons and dendrites.
■ The process of transmission is by diffusion of chemicals called
neuro-transmitters.
McCulloch-Pitts introduced a simplified model of this real neurons.
08

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
1.4 Artificial Neuron Model
An artificial neuron is a mathematical function conceived as a simple
model of a real (biological) neuron.
• The McCulloch-Pitts Neuron
This is a simplified model of real neurons, known as a Threshold Logic Unit.
Input1
Input 2
Input n
■ A set of input connections brings in activations from other neurons.
■ A processing unit sums the inputs, and then applies a non-linear
activation function (i.e. squashing / transfer / threshold function).
■ An output line transmits the result to other neurons.
In other words ,
- The input to a neuron arrives in the form of signals.
- The signals build up in the cell.
- Finally the cell discharges (cell fires) through the output .
- The cell can start building up signals again.
09
Σ Output

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
1.5 Notations
Recaps : Scalar, Vectors, Matrices and Functions
• Scalar : The number xi can be added up to give a scalar number.
s = x1 + x2 + x3 + . . . . + xn = xi
• Vectors : An ordered sets of related numbers. Row Vectors (1 x n)
X = ( x1 , x2 , x3 , . . ., xn ) , Y = ( y1 , y2 , y3 , . . ., yn )
Add : Two vectors of same length added to give another vector.
Z = X + Y = (x1 + y1 , x2 + y2 , . . . . , xn + yn)
Multiply: Two vectors of same length multiplied to give a scalar.
p = X . Y = x1 y1 + x2 y2 + . . . . + xnyn = xi yi
10
Σ
i=1
n
Σ
i=1
n

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Matrices : m x n matrix , row no = m , column no = n
w11 w11 . . . . w1n
w21 w21 . . . . w21
W = . . . . . . .
. . . . . . .
wm1 w11 . . . . wmn
Add or Subtract : Matrices of the same size are added or subtracted
component by component. A + B = C , cij = aij + bij
a11 a12 b11 b12 c11 = a11+b11 c12 = a12+b12
a21 a22 b21 b22 C21 = a21+b21 C22 = a22 +b22
Multiply : matrix A multiplied by matrix B gives matrix C.
(m x n) (n x p) (m x p)
elements cij = aik bkj
a11 a12 b11 b12 c11 c12
a21 a22 b21 b22 c21 c22
c11 = (a11 x b11) + (a12 x B21)
c12 = (a11 x b12) + (a12 x B22)
C21 = (a21 x b11) + (a22 x B21)
C22 = (a21 x b12) + (a22 x B22)
11
+ =
Σ
k=1
n
x =

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
1.6 Functions
The Function y= f(x) describes a relationship, an input-output mapping,
from x to y.
■ Threshold or Sign function : sgn(x) defined as
1 if x ≥ 0
sgn (x) =
0 if x < 0
Sign(x)
O/P
-4 -3 -2 -1 0 1 2 3 4 I/P
■ Threshold or Sign function : sigmoid(x) defined as a smoothed
(differentiable) form of the threshold function
1
sigmoid (x) =
1 + e
-x
Sign(x)
O/P
-4 -3 -2 -1 0 1 2 3 4 I/P
12
0
1
.2
.6
.4
.8
0
1
.2
.6
.8

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –Artificial Neuron Model
2. Model of Artificial Neuron
A very simplified model of real neurons is known as a Threshold Logic
Unit (TLU). The model is said to have :
- A set of synapses (connections) brings in activations from other neurons.
- A processing unit sums the inputs, and then applies a non-linear activation
function (i.e. squashing / transfer / threshold function).
- An output line transmits the result to other neurons.
2.1 McCulloch-Pitts (M-P) Neuron Equation
McCulloch-Pitts neuron is a simplified model of real biological neuron.
Input 1
Input 2
Input n
Simplified Model of Real Neuron
(Threshold Logic Unit)
The equation for the output of a McCulloch-Pitts neuron as a function
of 1 to n inputs is written as
Output = sgn ( Input i - Φ )
where Φ is the neuron’s activation threshold.
If Input i ≥ Φ then Output = 1
If Input i < Φ then Output = 0
In this McCulloch-Pitts neuron model, the missing features are :
- Non-binary input and output,
- Non-linear summation,
- Smooth thresholding,
- Stochastic, and
- Temporal information processing.
13
Σ Output
Σ
i=1
n
Σ
i=1
n
Σ
i=1
n

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
2.2 Artificial Neuron - Basic Elements
Neuron consists of three basic components - weights, thresholds, and a
single activation function.
Fig Basic Elements of an Artificial Linear Neuron
■ Weighting Factors w
The values w1 , w2 , . . . wn are weights to determine the strength of
input vector X = [x1 , x2 , . . . , xn]
T
. Each input is multiplied by the
associated weight of the neuron connection XT
W. The +ve weight
excites and the -ve weight inhibits the node output.
I = X
T
.W = x1 w1 + x2 w2 + . . . . + xnwn = xi wi
■ Threshold Φ
The node’s internal threshold Φ is the magnitude offset. It affects the
activation of the node output y as:
Y = f (I) = f { xi wi - Φk }
To generate the final output Y , the sum is passed on to a non-linear
filter f called Activation Function or Transfer function or Squash function
which releases the output Y.
14
W1
Σ
W2
Wn
x1
x2
xn
Activation
Function
i=1
Synaptic Weights
Φ
Threshold
y
Σ
i=1
n
Σ
i=1
n

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
■ Threshold for a Neuron
In practice, neurons generally do not fire (produce an output) unless
their total input goes above a threshold value.
The total input for each neuron is the sum of the weighted inputs
to the neuron minus its threshold value. This is then passed through
the sigmoid function. The equation for the transition in a neuron is :
a = 1/(1 + exp(- x)) where
x = ai wi - Q
a is the activation for the neuron
ai is the activation for neuron i
wi is the weight
Q is the threshold subtracted
■ Activation Function
An activation function f performs a mathematical operation on the
signal output. The most common activation functions are:
- Linear Function,
- Piecewise Linear Function,
- Tangent hyperbolic function
- Threshold Function,
- Sigmoidal (S shaped) function,
The activation functions are chosen depending upon the type of
problem to be solved by the network.
15
Σ
i

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network – Artificial Neuron Model
2.2 Activation Functions f - Types
Over the years, researches tried several functions to convert the input into
an outputs. The most commonly used functions are described below.
- I/P Horizontal axis shows sum of inputs .
- O/P Vertical axis shows the value the function produces ie output.
- All functions f are designed to produce values between 0 and 1.
• Threshold Function
A threshold (hard-limiter) activation function is either a binary type or
a bipolar type as shown below.
binary threshold
O/p
I/P
Output of a binary threshold function produces :
1 if the weighted sum of the inputs is +ve,
0 if the weighted sum of the inputs is -ve.
1 if I ≥ 0
Y = f (I) =
0 if I < 0
bipolar threshold
O/p
I/P
Output of a bipolar threshold function produces :
1 if the weighted sum of the inputs is +ve,
-1 if the weighted sum of the inputs is -ve.
1 if I ≥ 0
Y = f (I) =
-1 if I < 0
Neuron with hard limiter activation function is called McCulloch-Pitts model.
16
1
1
-1

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Piecewise Linear Function
This activation function is also called saturating linear function and can
have either a binary or bipolar range for the saturation limits of the output.
The mathematical model for a symmetric saturation function is described
below.
Piecewise Linear
O/p
I/P
This is a sloping function that produces :
-1 for a -ve weighted sum of inputs,
1 for a +ve weighted sum of inputs.
∝ I proportional to input for values between +1
and -1 weighted sum,
1 if I ≥ 0
Y = f (I) = I if -1 ≥ I ≥ 1
-1 if I < 0
17
+1
-1

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network – Artificial Neuron Model
• Sigmoidal Function (S-shape function)
The nonlinear curved S-shape function is called the sigmoid function.
This is most common type of activation used to construct the neural
networks. It is mathematically well behaved, differentiable and strictly
increasing function.
Sigmoidal function A sigmoidal transfer function can be
written in the form:
1
Y = f (I) = , 0 ≤ f(I) ≤ 1
1 + e -α I
= 1/(1 + exp(-α I)) , 0 ≤ f(I) ≤ 1
This is explained as
≈ 0 for large -ve input values,
1 for large +ve values, with
a smooth transition between the two.
α is slope parameter also called shape
parameter; symbol the λ is also used to
represented this parameter.
The sigmoidal function is achieved using exponential equation.
By varying α different shapes of the function can be obtained which
adjusts the abruptness of the function as it changes between the two
asymptotic values.
18
1 O/P
0.5
I/P
-4 -2 0 1 2
α = 1.0
α = 0.5
α = 2.0

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Example :
The neuron shown consists of four inputs with the weights.
Fig Neuron Structure of Example
The output I of the network, prior to the activation function stage, is
+1
+1
I = XT
. W = 1 2 5 8 = 14
-1
+2
= (1 x 1) + (2 x 1) + (5 x -1) + (8 x 2) = 14
With a binary activation function the outputs of the neuron is:
y (threshold) = 1;
19
+1
Σ
+1
+2
-1
x1=1
x2=2
xn=8
Activation
Function
Summing
Junction
Synaptic
Weights
Φ = 0
Threshold
y
X3=5
I

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network – Architecture
3. Neural Network Architectures
An Artificial Neural Network (ANN) is a data processing system, consisting
large number of simple highly interconnected processing elements as
artificial neuron in a network structure that can be represented using a
directed graph G, an ordered 2-tuple (V, E) , consisting a set V of vertices
and a set E of edges.
- The vertices may represent neurons (input/output) and
- The edges may represent synaptic links labeled by the weights attached.
Example :
Fig. Directed Graph
Vertices V = { v1 , v2 , v3 , v4, v5 }
Edges E = { e1 , e2 , e3 , e4, e5 }
20
V1 V3
V2 V4
V5
e3
e2
e5
e4
e5

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
3.1 Single Layer Feed-forward Network
The Single Layer Feed-forward Network consists of a single layer of
weights , where the inputs are directly connected to the outputs, via a
series of weights. The synaptic links carrying weights connect every input
to every output , but not other way. This way it is considered a network of
feed-forward type. The sum of the products of the weights and the inputs
is calculated in each neuron node, and if the value is above some threshold
(typically 0) the neuron fires and takes the activated value (typically 1);
otherwise it takes the deactivated value (typically -1).
Fig. Single Layer Feed-forward Network
21
w21
w11
w12
wn2
wn1
w1m
w2m
wnm
w22
y1
y2
ym
x1
x2
xn
output yj
input xi weights wij
Single layer
Neurons

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
3.2 Multi Layer Feed-forward Network
The name suggests, it consists of multiple layers. The architecture of
this class of network, besides having the input and the output layers,
also have one or more intermediary layers called hidden layers. The
computational units of the hidden layer are known as hidden neurons.
Fig. Multilayer feed-forward network in (ℓ – m – n) configuration.
- The hidden layer does intermediate computation before directing the
input to output layer.
- The input layer neurons are linked to the hidden layer neurons; the
weights on these links are referred to as input-hidden layer weights.
- The hidden layer neurons and the corresponding weights are referred to
as output-hidden layer weights.
- A multi-layer feed-forward network with ℓ input neurons, m1 neurons in
the first hidden layers, m2 neurons in the second hidden layers, and n
output neurons in the output layers is written as (ℓ - m1 - m2 – n ).
The Fig. above illustrates a multilayer feed-forward network with a
configuration (ℓ - m – n).
22
w11
w12
v21
v11
w1m
vn1
v1m
v2m
Vℓm
w11
x1
x2
xℓ
y3
y1
y2
yn
y1
ym
Hidden Layer
neurons yj
Output Layer
neurons zk
Input Layer
neurons xi
Input
hidden layer
weights vij
Output
hidden layer
weights wjk

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
3.3 Recurrent Networks
The Recurrent Networks differ from feed-forward architecture. A Recurrent
network has at least one feed back loop.
Example :
Fig Recurrent Neural Network
There could be neurons with self-feedback links; that is the output of a
neuron is fed back into it self as input.
23
x1
x2
Xℓ
y2
y1
Yn
y1
ym
Hidden Layer
neurons yj
Output Layer
neurons zk
Input Layer
neurons xi
Feedback
links

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –Learning methods
4. Learning Methods in Neural Networks
The learning methods in neural networks are classified into three basic types :
- Supervised Learning,
- Unsupervised Learning and
- Reinforced Learning
These three types are classified based on :
- presence or absence of teacher and
- the information provided for the system to learn.
These are further categorized, based on the rules used, as
- Hebbian,
- Gradient descent,
- Competitive and
- Stochastic learning.
24

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Classification of Learning Algorithms
Fig. below indicate the hierarchical representation of the algorithms
mentioned in the previous slide. These algorithms are explained in
subsequent slides.
Fig. Classification of learning algorithms
25
Neural Network
Learning algorithms
Unsupervised Learning
Supervised Learning
(Error based)
Reinforced Learning
(Output based)
Error Correction
Gradient descent
Stochastic
Back
Propagation
Least Mean
Square
Hebbian Competitive

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Supervised Learning
- A teacher is present during learning process and presents expected
output.
- Every input pattern is used to train the network.
- Learning process is based on comparison, between network's computed
output and the correct expected output, generating "error".
- The "error" generated is used to change network parameters that result
improved performance.
• Unsupervised Learning
- No teacher is present.
- The expected or desired output is not presented to the network.
- The system learns of it own by discovering and adapting to the structural
features in the input patterns.
• Reinforced learning
- A teacher is present but does not present the expected or desired output
but only indicated if the computed output is correct or incorrect.
- The information provided helps the network in its learning process.
- A reward is given for correct answer computed and a penalty for a wrong
answer.
Note : The Supervised and Unsupervised learning methods are most popular
forms of learning compared to Reinforced learning.
26

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Hebbian Learning
Hebb proposed a rule based on correlative weight adjustment.
In this rule, the input-output pattern pairs (Xi , Yi) are associated by
the weight matrix W, known as correlation matrix computed as
W = Xi YiT
where Yi
T
is the transpose of the associated output vector Yi
There are many variations of this rule proposed by the other
researchers (Kosko, Anderson, Lippman) .
27
Σ
i=1
n

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Gradient descent Learning
This is based on the minimization of errors E defined in terms of weights
and the activation function of the network.
- Here, the activation function of the network is required to be
differentiable, because the updates of weight is dependent on
the gradient of the error E.
- If ∆ Wij is the weight update of the link connecting the i th and the j th
neuron of the two neighboring layers, then ∆ Wij is defined as
∆ Wij = η (∂ E / ∂ Wij )
where η is the learning rate parameters and (∂ E / ∂ Wij ) is error
gradient with reference to the weight Wij .
Note : The Hoffs Delta rule and Back-propagation learning rule are
the examples of Gradient descent learning.
28

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Competitive Learning
- In this method, those neurons which respond strongly to the input
stimuli have their weights updated.
- When an input pattern is presented, all neurons in the layer compete,
and the winning neuron undergoes weight adjustment .
- This strategy is called "winner-takes-all".
• Stochastic Learning
- In this method the weights are adjusted in a probabilistic fashion.
- Example : Simulated annealing which is a learning mechanism
employed by Boltzmann and Cauchy machines.
29

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –Systems
5. Taxonomy Of Neural Network Systems
In the previous sections, the Neural Network Architectures and the
Learning methods have been discussed. Here the popular neural network
systems are listed. The grouping of these systems in terms of architectures
and the learning methods are presented in the next slide.
• Neural Network Systems
– ADALINE (Adaptive Linear Neural Element)
– ART (Adaptive Resonance Theory)
– AM (Associative Memory)
– BAM (Bidirectional Associative Memory)
– Boltzmann machines
– BSB ( Brain-State-in-a-Box)
– Cauchy machines
– Hopfield Network
– LVQ (Learning Vector Quantization)
– Neoconition
– Perceptron
– RBF ( Radial Basis Function)
– RNN (Recurrent Neural Network)
– SOFM (Self-organizing Feature Map)
30

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –Systems
• Classification of Neural Network
A taxonomy of neural network systems based on Architectural types
and the Learning methods is illustrated below.
Learning Methods
Gradient
descent
Hebbian Competitive Stochastic
Single-layer
feed-forward
ADALINE,
Hopfield,
Percepton,
AM,
Hopfield,
LVQ,
SOFM
-
Multi-layer
feed- forward
CCM,
MLFF,
RBF
Neocognition
Recurrent
Networks
RNN BAM,
BSB,
Hopfield,
ART Boltzmann and
Cauchy
machines
Table : Classification of Neural Network Systems with respect to
learning methods and Architecture types
31

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –Single Layer learning
6. Single-Layer NN Systems
Here, a simple Perceptron Model and an ADALINE Network Model is presented.
6.1 Single layer Perceptron
Definition : An arrangement of one input layer of neurons feed forward
to one output layer of neurons is known as Single Layer Perceptron.
Fig. Simple Perceptron Model
1 if net j ≥ 0
y j = f (net j) = where net j = xi wij
0 if net j < 0
32
Σ
i=1
n
w21
w11
w12
wn2
wn1
w1m
w2m
wnm
w22
y1
y2
ym
x1
x2
xn
output yj
input xi weights wij
Single layer
Perceptron

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Learning Algorithm : Training Perceptron
The training of Perceptron is a supervised learning algorithm where
weights are adjusted to minimize error when ever the output does
not match the desired output.
− If the output is correct then no adjustment of weights is done.
i.e. =
− If the output is 1 but should have been 0 then the weights are
decreased on the active input link
i.e. = − α . xi
− If the output is 0 but should have been 1 then the weights are
increased on the active input link
i.e. = + α . xi
Where
is the new adjusted weight, is the old weight
xi is the input and α is the learning rate parameter.
α small leads to slow and α large leads to fast learning.
33
W
i j
K+1
W
i j
K+1
W
i j
K
W
i j
K+1
W
i j
K
W
i j
K+1
W
i j
K
W
i j
K

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Perceptron and Linearly Separable Task
Perceptron can not handle tasks which are not separable.
- Definition : Sets of points in 2-D space are linearly separable if the
sets can be separated by a straight line.
- Generalizing, a set of points in n-dimensional space are linearly
separable if there is a hyper plane of (n-1) dimensions separates
the sets.
Example
S1 S2 S1
S2
(a) Linearly separable patterns (b) Not Linearly separable patterns
Note : Perceptron cannot find weights for classification problems that
are not linearly separable.
34

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• XOR Problem :
Exclusive OR operation
Input x1 Input x2 Output
0 0 0
1 1 0
0 1 1
1 0 1
XOR truth table
X2
(0, 1) (1, 1)
(0, 0) X1
(0, 1)
Fig. Output of XOR in
X1 , x2 plane
Even parity is, even number of 1 bits in the input
Odd parity is, odd number of 1 bits in the input
- There is no way to draw a single straight line so that the circles are on
one side of the line and the dots on the other side.
- Perceptron is unable to find a line separating even parity input
patterns from odd parity input patterns.
35
•
°
•
°
Even parity •
Odd parity °

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
• Perceptron Learning Algorithm
The algorithm is illustrated step-by-step.
■ Step 1 :
Create a peceptron with (n+1) input neurons x0 , x1 , . . . . . , . xn ,
where x0 = 1 is the bias input.
Let O be the output neuron.
■ Step 2 :
Initialize weight W = (w0 , w1 , . . . . . , . wn ) to random weights.
■ Step 3 :
Iterate through the input patterns Xj of the training set using the
weight set; ie compute the weighted sum of inputs net j = xi wi
for each input pattern j .
■ Step 4 :
Compute the output y j using the step function
1 if net j ≥ 0
y j = f (net j) = where net j = xi wij
0 if net j < 0
■ Step 5 :
Compare the computed output yj with the target output yj for
each input pattern j .
If all the input patterns have been classified correctly, then output
(read) the weights and exit.
■ Step 6 :
Otherwise, update the weights as given below :
If the computed outputs yj is 1 but should have been 0,
Then wi = wi - α xi , i= 0, 1, 2, . . . . , n
If the computed outputs yj is 0 but should have been 1,
Then wi = wi + α xi , i= 0, 1, 2, . . . . , n
where α is the learning parameter and is constant.
■ Step 7 :
goto step 3
■ END
36
Σ
i=1
n
Σ
i=1
n

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –ADALINE
6.2 ADAptive LINear Element (ADALINE)
An ADALINE consists of a single neuron of the McCulloch-Pitts type,
where its weights are determined by the normalized least mean
square (LMS) training law. The LMS learning rule is also referred to as
delta rule. It is a well-established supervised training method that
has been used over a wide range of diverse applications.
• Architecture of a simple ADALINE
The basic structure of an ADALINE is similar to a neuron with a
linear activation function and a feedback loop. During the training
phase of ADALINE, the input vector as well as the desired output
are presented to the network.
[The complete training mechanism has been explained in the next slide. ]
37
W1
Σ
W2
Wn
Σ
x1
x2
xn
Neuron
Error
Desired Output
Output
–
+

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –ADALINE
• ADALINE Training Mechanism
(Ref. Fig. in the previous slide - Architecture of a simple ADALINE)
■ The basic structure of an ADALINE is similar to a linear neuron
with an extra feedback loop.
■ During the training phase of ADALINE, the input vector
X = [x1 , x2 , . . . , xn]
T
as well as desired output are presented
to the network.
■ The weights are adaptively adjusted based on delta rule.
■ After the ADALINE is trained, an input vector presented to the
network with fixed weights will result in a scalar output.
■ Thus, the network performs an n dimensional mapping to a
scalar value.
■ The activation function is not used during the training phase.
Once the weights are properly adjusted, the response of the
trained unit can be tested by applying various inputs, which are
not in the training set. If the network produces consistent
responses to a high degree with the test inputs, it is said
that the network could generalize. The process of training and
generalization are two important attributes of this network.
Usage of ADLINE :
In practice, an ADALINE is used to
- Make binary decisions; the output is sent through a binary threshold.
- Realizations of logic gates such as AND, NOT and OR .
- Realize only those logic functions that are linearly separable.
38

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –Applications
7. Applications of Neural Network
Neural Network Applications can be grouped in following categories:
■ Clustering:
A clustering algorithm explores the similarity between patterns and
places similar patterns in a cluster. Best known applications include
data compression and data mining.
■ Classification/Pattern recognition:
The task of pattern recognition is to assign an input pattern
(like handwritten symbol) to one of many classes. This category
includes algorithmic implementations such as associative memory.
■ Function approximation :
The tasks of function approximation is to find an estimate of the
unknown function subject to noise. Various engineering and scientific
disciplines require function approximation.
■ Prediction Systems:
The task is to forecast some future values of a time-sequenced
data. Prediction has a significant impact on decision support systems.
Prediction differs from function approximation by considering time factor.
System may be dynamic and may produce different results for the
same input data based on system state (time).
39

R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
R
C
C
h
a
k
r
a
b
o
r
t
y
,
w
w
w
.
m
y
r
e
a
d
e
r
s
.
i
n
f
o
SC - Neural Network –References
8. References : Textbooks
1. "Neural Network, Fuzzy Logic, and Genetic Algorithms - Synthesis and
Applications", by S. Rajasekaran and G.A. Vijayalaksmi Pai, (2005), Prentice Hall,
Chapter 2, page 11-33.
2. "Soft Computing and Intelligent Systems Design - Theory, Tools and Applications",
by Fakhreddine karray and Clarence de Silva (2004), Addison Wesley, chapter 4,
page 223-248.
3. "Neural Networks: A Comprehensive Foundation", by Simon S. Haykin, (1999),
Prentice Hall, Chapter 1-7, page 1-363.
4. "Elements of Artificial Neural Networks", by Kishan Mehrotra, Chilukuri K. Mohan
and Sanjay Ranka, (1996), MIT Press, Chapter 1-5, page 1-214.
5. "Fundamentals of Neural Networks: Architecture, Algorithms and Applications", by
Laurene V. Fausett, (1993), Prentice Hall, Chapter1-4, page 1-214.
6. "Neural Network Design", by Martin T. Hagan, Howard B. Demuth and Mark
Hudson Beale, ( 1996) , PWS Publ. Company, Chapter 1-7, page 1-1 to 7-31.
7. "An Introduction to Neural Networks", by James A. Anderson, (1997), MIT Press,
Chapter 1- 12, page 1-401.
8. Related documents from open source, mainly internet. An exhaustive list is
being prepared for inclusion at a later date.
40

Fundamentals of Neural Network (Soft Computing)

Recommended

More Related Content

What's hot (20)

Similar to Fundamentals of Neural Network (Soft Computing) (20)

More from Amit Kumar Rathi (20)

Recently uploaded (20)

Fundamentals of Neural Network (Soft Computing)