You are on page 1of 10

Artificial Neural Networks

 Artificial Neural Networks(ANN) is a mathematical


model, inspired by the biological neural network.
 In a ANN processing element called neurons are
connected together to form a network of nodes
mimicking the biological neural network.
 A ANN changes its structure through information
that flows through the network during the learning
phase.
Continued..
 A ANN is a adaptive system.
 The error information is feedback to the
system which makes all adjustment to their
parameters in a systematic fashion.
 For the simple neuron network architecture
the net input has to be calculated as in the
following ways:
yin=-x1w1+x2w2
Where x1 & x2 are the activation of the input
neurons X1 and X2.
Output=Function(net input calculated)
y=f(yin)
Activation Function

It is used to combine activation information from an external source to provide


output.
Activation function may be:
Identity function : f(x)=x.
1 x>=$
Binary step function : f(x)=
0 x<$

+1 x>=$
Bipolar step function : f(x)=
-1 x<$

Sigmodial function

$ represents the threshold value.


Biological model consist of :-

 Cell body or Soma: where the cell nucleus is located.


 Dendrites: where the nerve is connected to the cell
body.
 Axon: which carries the impulses of neuron.
Artificial neural network model
M-P neuron model.
• Activation is binary.
• If yin >=threshold=>neuron fires.

Hebb neuron model.


• Learning performed by change in synaptic
gap.
• In this bipolar function is prefered.
Types of ANN
 Single layered feed-forward network.
 Multilayer feed-forward network.
Training of Artificial Neural Networks(Learning process)
Supervised Learning

Neural
X
Network Y
(Actual Output)
(Input) W
Error signal D
Error(D-Y)
Unsupervised learning signals generator Desired output

ANN Y
X
W
Reinforcement Learning

Neural
X
Network Y

(Input) W (Actual Output)

Error signal R
Error
signals
generator Reinforcement
Signal

Supervise learning supports back


propagation network.

•By this method error is back propagated to the hidden


layer.
Advantages:
1: When element of the neural network fails, it can continue
without any problem by their parallel nature.
2: A neural network learns and doesn’t need to be
reprogrammed.

Disadvantages:
1:It requires high processing time for large neural networks.
2.Neural network needs training to operate.

You might also like