A Survey On Genetic Algorithms

Shreeraj Pawar. International Journal of Engineering Research and Applications
www.ijera.com
ISSN: 2248-9622, Vol. 10, Issue 12, (Series-III) December 2020, pp. 25-29
www.ijera.com DOI: 10.9790/9622-1012032529 25 | P a g e
A Survey on Genetic Algorithms
Shreeraj Pawar
*(Department of Computer Engineering, Fr. C. Rodrigues Institute of Technology, Vashi-400703.
ABSTRACT
This is a summary paper of a simple genetic algorithm. A genetic algorithm is used to find a way out of them
maze generated by walls. Here a player is simulated in a population, and then evolved to further as per the fitness
of each player. The child with the highest fitness is carried on the next generation and is breed to give a new
population. This paper covers the applications of genetic algorithms and how it was used to simulate a player
running through a maze of walls trying to reach its goal and few applications of genetic algorithms.
Keywords – Artificial intelligence, Genetic algorithm, generation, fitness
---------------------------------------------------------------------------------------------------------------------------------------
Date of Submission: 08-12-2020 Date of Acceptance: 24-12-2020
---------------------------------------------------------------------------------------------------------------------------------------
I. INTRODUCTION
Evolution is the change in characteristics of
biological population over successive generations.
These characteristics are the expressions of genes
that are passed on from parent to offspring during
reproduction. The process of evolution was vastly
explained by Sir Charles Darwin in his book “On the
Origin of Species” in 1859. Darwin in his theory
later known - Darwin’s Theory of Evolution, says
that evolution by natural selection is the process by
which organisms change over time as a result of
changes in heritable physical or behavioral traits.
Changes that allow an organism to better adapt to its
environment will help it survive and this benefits the
future generations. Evolution by natural selection is
one of the best substantiated theories in the history
of science, supported by evidence from a wide
variety of scientific disciplines like geology,
genetics and developmental biology.
In computer science, evolutionary
computation is a subfield of artificial intelligence
and soft computing inspired by biological evolution.
In technical terms, they are a family of population-
based trial and error problem solvers with stochastic
optimization character. Here the initial set of
candidates are generated and iteratively updated.
Each new generation is produced by removing less
desired solutions, and introducing small random
changes. In biological terminology, a population of
solutions is subjected to natural selection (or
artificial selection) and mutation. As a result, the
population will gradually evolve to increase in
fitness. Evolutionary algorithms form a subset of
evolutionary computation which involve techniques
implementing mechanisms inspired by biological
evolution such as reproduction, mutation,
recombination, natural selection and survival of the
fittest.
Evolutionary computing techniques mostly
involve optimization algorithms which are designed
to find, generate or select a heuristic that may
provide sufficiently good solution. In a broad sense
we are talking about fields like:
 Genetic Algorithms
 Genetic Programming
 Neuroevolution
 Ant Colony Optimization
 Artificial Immune Systems
 Swarm intelligence
II. GENETIC ALGORITHM
In Computer science, a Genetic algorithm
(GA) is a randomized search and optimization
technique inspired by natural selection that belongs
to a larger class of Evolutionary algorithms. The GA
was introduced by John Holland in 1975. He was
inspired by the concept of Darwinian’s principle of
survival of the fittest individuals and natural
selection and developed the theory of genetic
algorithm. They are categorized as global search
heuristics as they give a optimal global solution to a
problem. In genetic algorithm, a population of
candidate solutions called individuals or phenotypes
are evolved towards a better solution. Each
candidate solution has a set of properties (its
chromosomes or genotype) which can be mutated
and altered. Every genetic algorithm has a few
components which are as follows:
A. Generation
Evolution starts with a population of
randomly generated individuals and is an iterative
process with population of each iteration is called as
RESEARCH ARTICLE OPEN ACCESS

www.ijera.com
Generation. A generation can of any size depending
on the computing capabilities of the system the
algorithms will be running on. For our Maze runner I
created a population of 500 candidates, thus each
generation has 500 individuals with completely
random set of chromosomes. Before we head
further, we need to understand terminologies like
genotype and phenotype.
• Genotype: A genotype is an organism’s complete
set of heritable genes or genes that can be passed
down from parent to offspring. These genes help
encode the characteristics that are physically
expressed in an organism such as hair color, height,
weight, etc.
• Phenotype: A phenotype is the term used in
genetics for the composite observable characteristics
or traits of an organism. The term covers organism’s
morphology or physical features, structure, its
developmental processes, its behavior, etc.
Our program has candidates which as essentially
dots on a screen. These dots represent an array of
randomly generated vectors which tell the dot the
direction in which it will move. So, our phenotype is
limited to only the shape, size and color of the dot,
whereas the genotype consists of 3 arrays of long set
of directional vectors which signify position,
velocity and acceleration.
B. Fitness Function
A fitness function is a particular type of
objective function that is used. A figure of merit is a
quantity used to characterized the performance of a
device, system or method relative to its alternatives.
The fitness function tells the algorithm how close a
particular individual is to achieving the desired
outcome. Fitness functions are used in genetic
programming and genetic algorithms to guide
simulations towards optimal design solutions.
When a population is created it contains
individuals or candidates with completely random
set of genes. The population is simulated to do a task
in our case traverse the maze without dying and
reach the goal. The dots traverse the maze with
directions corresponding to the vectors generated
during initialization. Fitness is given by the inverse
of the square of the distance from its goal. It is as
follows:
fitness=1/( distanceToGoal *
distanceToGoal)
The fitness function changes when a
candidate or individual of a generation reaches the
goal. The fitness function is then as follows:
fitness =
Here ‘step’ is the number of steps taken by
an individual before dying. Each step is incremented
based on the velocity and acceleration of each dot.
Constants like 16 and 1000 act as hyperparameters
which were tuned during testing of the fitness
function.
C. Selection
Natural selection is a key mechanism in
which only the strongest genes are selected and
passed on to the next generation. During an
individual’s lifespan based on environmental factors
some genes get suppressed and some get activated to
a large extend. The activated genes are key to
survival of a species and are then passed on to the
next generation. These new genes give rise to
different characteristics or phenotype in the next
generation.
Selection is the stage of genetic algorithm
in which individual genomes are chosen from a
population for later breeding. Selection is done
based on the fitness calculated by the fitness
function. The top 2 individuals are selected as the
best candidates for gene crossover. There are many
methods to Selections namely Roulette Wheel
Selection, Rank Selection, Steady State Selection,
Tournament Selection, Elitism Selection, Boltzmann
Selection. For our Maze Runner problem, we used
Elitism Selection where the best individual from last
generation is carried over to the next one.
D. Crossover and Mutation
In genetic algorithms and evolutionary
computation, crossover, also called recombination is
a function where genetic information of two or more
individuals is used to generate a new offspring. Our
Maze runner Dot contains various vectors for
position, velocity, and acceleration. The directions
array at the initialization consists of random angles
from 0 to 2π. After the end of generation 1 that is
when all the individuals of the generation die the
fitness is calculated for each individual and the top 2
are selected. The best individual is simply carried on
to the next generation and the 2nd best is cross-
overed with the best one. Here first half of the best
individual’s genome is taken and combined with the
second half of the 2nd best candidate. Thus, the
resultant genome is a recombination of the best and
2nd best candidates of the previous generation.
Mutation is the alteration of a genome of an
organism. It alters one or more genes thus generation
variation within a species. Variations can be subtle
or may entirely change the characteristics of an
organism from its parent. In genetic algorithms,
mutation is introduced to bring about some genetic
diversity from one generation to next. The mutation
within a generation is governed by a parameter
called the rate of mutation or mutation rate. It is the
probability of whether a given candidate will
undergo mutation or not. In our Maze runner,

www.ijera.com
mutation occurs when a random generated variable
‘n’ is less than the mutation rate. If so then all the
values in the directions array are added with the
random number ‘n’. Mutation rate should be very
low as we don’t want our new generation to be
completely different than this previous as this may
lead to diverting from the main goal.
III. APPLICATIONS
Genetic Algorithms can be used to solve
many problems which traditional search algorithms
would take many hours or days or years to get to the
optimal solution. Some application of genetic
algorithms are as follows:
1)NEAT Algorithm
NEAT is an acronym for NeuroEvolution
of Augmenting Topologies is an algorithm or a
method of optimizing weights that are used in neural
networks.
Weights in a neural network are randomly initialized
and then changed after each epoch. Using a loss
function and learning rate the weights are adjusted
so as to give correct output for a given input.
Genetic algorithm can be used to optimize the
weights on between connections and significantly
enhance the performance of the NN. Here a
population is generated with each candidate having a
NN with randomized weights. After a generation the
candidates with the maximum fitness or the best
performing candidates of a generation are cross
overed and mutated. During crossover the weights
are recombined and not the input and output. Thus,
over generations the weights get optimized such that
the NN always reaches the desired output. Mutation
in NEAT can change both connection weights and
network structures. Connection weights mutate as in
any NE system, with each connection either
perturbed or not at each generation. New nodes may
also get added to the NN and the resultant NN after
several generation may look completely different
than its initialization. NEAT is a powerful method
for artificially evolving neural networks. It
demonstrates evolving topology along with weights
can be a major advantage.
2)Solving Travelling Salesman Problem
The Travelling Salesman problem (TSP) is
one of the old problems in computer science. Given
a network with ‘n’ nodes (cities), from a starting
node say ‘S’ what path should be taken such that all
the nodes can be visited with least amount of cost
(cost here can be distance or time). The problem is
to find the least cost Hamiltonian cycle. The number
of ways which gives minimum cost can increases
exponentially with more number of cities. Many
real-world problems can be modelled after TSP and
thus has attracted many researchers and
mathematicians. Finding the optimal solution
involves evaluating every possible Hamiltonian path,
but this is highly inefficient even for a moderate size
graph. Since practical applications require solving
larger problems, hence emphasis has shifted from
the aim of finding exactly optimal solutions to TSP,
to the aim of getting, heuristically, ‘good solutions’
in reasonable time. Genetic algorithm (GA) is one of
the best heuristic algorithms that have been used
widely to solve the TSP instances. Many crossover
operators and mutation operations have been
implemented that give a heuristically ‘good solution’
and traditional search algorithms.
3)Signal Processing
Genetic Algorithms being a powerful
optimization tool have a large number of
applications in signal processing. Signal processing
involves noise cancellation, time-delay estimation,
etc. GA are well suited for techniques like Adaptive
filtering where filter coefficients are needed to be
optimized. Gas can be applied to solve time-delay
problems where any modeling of gain factors and
delay is not required and direct estimates of both
parameters are obtained through the genetic
optimization procedure. GA also can be used for
Active noise cancellation (ANC). ANC is a
technique that uses secondary sound sources to
generate sound waves that create destructive
interference canceling the undesired sound. GA can
be applied to find optimally controlled secondary
sound signals such that minimum noise is perceived.
IV. OBSERVATIONS
The entire project was made in a sketching
tool called processing3. Processing is a JavaScript
based sketching tool used for creative coding. The
project consists of a window with a starting point of
the player and a goal point. The path between is
made up of walls both horizontal and vertical. If the
player hits the window boundary or any wall, the
player dies. When we start the program, in
generation 1 as the vectors are randomly assigned,
the players move around randomly and die quicky.
After generation 15, the player moves in a particular
path and less player die from random movements.
At generation 32 the player always avoids
the initial walls and boundaries and reach near the
goal. It is until generation 38 that one of the
candidates in the population reaches the goal. Now
the fitness function changes as described earlier and
we seen a large number of candidates reaching the
goal. By generation 44, 425 out of 500 reach the
goal avoiding all the obstacles. I kept the algorithm
running and later found a limitation or a bug within
my program which is covered in the next section. It
can be seen in the following graph fig1.0 that after
generation 55 the number of candidates reaching the

www.ijera.com
goal does not improve but reduces with increasing
generations.
Fig1.0: Line 1 indicates the
number of candidates
V. LIMITATIONS
Although evolutionary approach and
genetic algorithms seem like a faster alternative to
traditional search algorithms, they have their own set
of limitations. Due to the randomness of the GA
operation, it is difficult to predict its performance
and time of operation which are essential factors for
hard-deadline real-time applications.
Finding the right fitness function can be
challenging at times. It may take hours or days to
tweak the fitness functions hyperparameters to find
the optimum fitness of each individual. Moreover, if
large population exits calculating fitness of each
individual can be computationally heavy, so efficient
functions needs to be programmed.
Genetic algorithms can be endless as in
they will keep on running until we manually stop
them. This may lead to converging towards a local
optimum rather than a global optimum. During our
testing we found that after generation 50 the dots
used to take a path which led to the death of many
candidates as compared to previous generations. Due
to constant mutation after every generation even the
best dot who took the shortest path in generation 44
to reach the goal died in the next generation. Since a
stopping condition was not added the algorithm kept
on running. The global maxima as achieved when a
dot takes the shortest path which in a rectangular
window is the diagonal. As there are obstacles in the
path candidates simply can’t traverse diagonally
hence maximum fitness could not have been
achieved.
VI. CONCLUSION
Thus, genetic approach can be seen as a
viable approach towards an underlying problem.
Genetic algorithms can be used for various other
applications and due to their versatility, they are
used to find optimum solutions of problems across
various fields.
Evolutionary computation is ought to improve with
advancements in parallel computing and by
developing more efficient algorithms.
ACKNOWLEDEMENTS
This paper is based on a project I
participated in during our 2nd year of our bachelor’s
degree. I would like to thank my professors who
encouraged me to write a research paper based on
the project. This has been an educational journey
and it would not have been possible without the
constant support from my mentors.
REFERENCES
[1] S. Forrest. Genetic algorithms: Principles of
adaptation applied to computation. Science,
261:872{878, Aug. 13 1993.
[2] Bäck T. Optimal mutation rates in genetic
search. InProceedings of the fifth international
conference on genetic algorithms 1993.
[3] Janikow, C.Z. A Knowledge-Intensive Genetic
Algorithm for Supervised Learning. Machine
Learning 13, 189–228 (1993).
[4] de Jong, K., 1988. Learning with Genetic
Algorithms: An Overview. Machine Learning,
3(2), pp.121-138.
[5] M. Srinivas and L. M. Patnaik, "Genetic
algorithms: a survey," in Computer, vol. 27,
no. 6, pp. 17-26, June 1994, doi:
10.1109/2.294849.
[6] K. S. Tang, K. F. Man, S. Kwong and Q. He,
"Genetic algorithms and their applications," in
IEEE Signal Processing Magazine, vol. 13,
no. 6, pp. 22-37, Nov. 1996, doi:
10.1109/79.543973.
[7] Ahmed, Z.H., 2010. Genetic algorithm for the
traveling salesman problem using sequential
constructive crossover operator. International
Journal of Biometrics & Bioinformatics
(IJBB), 3(6), p.96.
[8] Tang, K.S., Man, K.F., Kwong, S. and He, Q.,
1996. Genetic algorithms and their
applications. IEEE signal processing
magazine, 13(6), pp.22-37.
[9] Stanley, K.O. and Miikkulainen, R., 2002.
Evolving neural networks through augmenting
topologies. Evolutionary computation, 10(2),
pp.99-127.
[10] Grefenstette, J., Gopal, R., Rosmaita, B. and
Van Gucht, D., 1985, July. Genetic algorithms
for the traveling salesman problem. In
Proceedings of the first International
Conference on Genetic Algorithms and their
Applications (Vol. 160, No. 168, pp. 160-168).
Lawrence Erlbaum.

www.ijera.com
[11] Koza, J.R. and Koza, J.R., 1992. Genetic
programming: on the programming of
computers by means of natural selection (Vol.
1). MIT press.
[12] Montana, D.J., 1995. Strongly typed genetic
programming. Evolutionary computation, 3(2),
pp.199-230.
[13] Schmitt, Lothar M. "Theory of genetic
algorithms." Theoretical Computer
Science 259, no. 1-2 (2001): 1-61.
[14] Clack C., Yu T. (1997) Performance enhanced
genetic programming. In: Angeline P.J.,
Reynolds R.G., McDonnell J.R., Eberhart R.
(eds) Evolutionary Programming VI. EP 1997.
Lecture Notes in Computer Science, vol 1213.
Springer, Berlin, Heidelberg.
Shreeraj Pawar. “A Survey on Genetic Algorithms.” International Journal of Engineering
Research and Applications (IJERA), vol.10 (12), 2020, pp 25-29.

A Survey On Genetic Algorithms

More Related Content

Similar to A Survey On Genetic Algorithms

More from Valerie Felton

Recently uploaded

A Survey On Genetic Algorithms