2. NEURAL NETWORKS USING GENETIC ALGORITHMS.pptx

NEURAL NETWORKS USING GENETIC
ALGORITHMS
RAJITHA NELLEPALLI
DEPARTMENT OF MASTER OF COMPUTER APPLICATIONS,SIETK
EMAIL ID:- rajithanellepalli18@gmail.com

ABSTRACT
• Combining neural network with evolutionary algorithms leads to
evolutionary artificial neural network. Evolutionary algorithms like GA
to train neural nets choose their structure or design related aspects
like the functions of their neurons. Along with basic concepts of
neural networks and genetic algorithm this paper includes a flexible
method for solving travelling salesman problem using genetic
algorithm. This offers a solution which includes a genetic algorithm
implementation in order to give a maximal approximation of the
problem with the reduction of cost.

KEYWORDS
 Genetic algorithm, Neural network, Travelling salesman problem

INTRODUCTION
• Genetic algorithm and neural networks are both inspired by
computation in biological system. A good deal of biological neural
architecture is determined genetically. Neural networks and genetic
algorithms are two techniques for optimization and learning, each
having its own strengths process tries to artificially reproduce the
mating process where the DNA of two parents determines the DNA
for the newly born.

Genetic algorithm :
“Genetic algorithms are based on mechanics of natural selection and
natural genetics”. This method combines Darwinian style survival of
the fittest among binary string "artificial creatures" with a structured,
yet randomized information exchange.

Fitness function(def)
In Genetic algorithms, each solution is generally represented as a string
of binary numbers, known as a chromosome.
Genetic algorithm have three main operations :-
 Reproduction(individual strings are copied according to their
fitness. that is having more chances to survive in next generation.)
 Crossover
 Mutation

CROSSOVER & MUTATION
• Here the crossover site is 7.
• After bit 7 the value of p-1 and p-2 get
interchanged, and the results as child 1
and child 2
• Mutation is carried out by flipping some digits
of a string which generates new solutions.
• Here mutation takes place at bit 7.
• Because of flip/swap the value of bit 7
changed from 1 to 0.

NEURAL NETWORK
• Let us review the basics of a neural networks . A neural network is a
Computational model consisting of a number of connected elements
knows as neurons.
• This neuron is a processing unit that receives input from outside the
network and /or from others neurons . Applies a local transformation
to that input and provides a single output signal which is passed on
to others neurons outside the network .

• The main elements or blocks of an artificial neural networks are as
follows:
a)The computing element
b)The connection pattern among the elements
c)The process used for training the neural network
An artificial neural network is composed of many artificial neurons that
are linked together according to specific network architecture. The
objective of the neural network is to transform the inputs into
meaningful outputs.

How to apply GA to Neural Networks
Combining Neural Nets with Evolutionary Algorithms leads to
Evolutionary Artificial Neural Networks (EANNs). One can use
Evolutionary Algorithms like the GA to train Neural Nets, choose their
structure or design related aspects like the function of their neurons.
Using GA to Train Neural Network:
First why one use GA to train Neural Networks: GA will train the
network no matter how it is connected - whether it’s a feed-forward or
a feedback network. Furthermore, it can train general networks which
are a mixture of the two types

Simple Neuro network String/Chromosome
chromosome
 All the weights in the network are
joined to make one string.
 The String is then used in the GA as a
member of the population. String
represents the weights of a complete
network.
• Depicts the value of chromosome
obtained from simple neural
network .

 How to evaluate Fitness
Fitness is measured by calculating the error (target – output) (i.e. fitness= 1/error) - the
lower the error the higher the fitness.
Example
The target for a network with a particular input is 1.The outputs are shown below ,calculate
their fitness.
Population
Member
1
2
3
4
Output
0.4
0.2
1.6
-0.9
Populatio
n
Members
1
2
3
4
Output
0.4
0.2
1.6
-0.9
Error
(T-0)
0.6
0.8
-0.6
1.9
Positive
0.6ss
0.8
0.6
1.9
Fitness
1.67
1.26
1.67
0.53

Implementation of GA in travelling
salesman problem
The genetic algorithms are more appropriately said to be an optimization technique
based on natural evolution. They include the survival of the fittest idea algorithm.
The idea is to first ‘guess’ the solutions and then combining the fittest solution to
create a new generation of solutions which should be better than the previous
generation I
 Methodology:
simple GA works by randomly generating an initial population of string ,which is referred as gene pool
and then apply
Operators to create new, and hopefully ,better population as successive generations.
 The first operator is reproduction where string are copied to the next generation with some probability
based on their
Objective functions value.
 The second operator is crossover where randomly selected pairs of strings are mated , creating new strings.
 The third operator ,mutation , is the occasional random alteration of the value at a string position.

Flow chart
In this flow chart we can see various steps
are following while implementing genetic
Algorithms to travelling salesman problem.
First create initial population, and then
evaluate fitness of all the chromosomes by
Applying fitness function to it.

 Application area of GA/ Neural network
a)Automotive Design
b)Robotics
C)Evolvable Hardware
d)Biomimetics Invention

Pros Cons
1. GA helps to generates better
population from good parents.
2. These results close to global
optimum.
3. Important character of GA , it is
robust.
4. They works well in various filed as:
In pattern matching
speech recognition , text-to-speech
Machines that are able learn
optical character recognition(OCR)
Fraudulent credit card detection(VISA)
1. It remains a “black box ” which
once fed with inputs produces an
output.
2. However ,their excellent result
record might compensate for that
deficiency.
3. A second drawback is that
inputs have to be altered before
being fed to the network.
It fails to depict following:
which network to use?
How many hidden layers?

Conclusion
This paper makes an effort to give a review with respect to
neural networks, genetic algorithm and how they both work
together. Genetic algorithm has three main operators: selection,
mutation and crossover. Neural network is computational model
having number of processing elements called neurons. These
techniques are black box which once fed with inputs produces an
output. As genetics and neural networks have a wide real-world
application area, also suffers from various cons so in future it will try
to work on these limitations. Reference

2. NEURAL NETWORKS USING GENETIC ALGORITHMS.pptx

More Related Content

What's hot (20)

Similar to 2. NEURAL NETWORKS USING GENETIC ALGORITHMS.pptx (20)

Recently uploaded (20)

2. NEURAL NETWORKS USING GENETIC ALGORITHMS.pptx