Sunteți pe pagina 1din 5

2010 Seventh International Conference on Information Technology

A Meiosis Genetic Algorithm

Nattavut Wiriyasermkul 1 Pisit Chanvarasuth


Veera Boonjing 2
School of Management Technology
School of Computer Science Sirindhorn International Institute of Technology,
King Mongkut's Institute of Technology Ladkrabang Pathumthani 12000, Thailand
Email : 1 nattavut_fee@yahoo.com e-mail: pisit@siit.tu.ac.th
2
kbveera@kmitl.ac.th
climbing and improving speed of optimal solution
Abstract-This paper proposes a new genetic algorithm convergence. In section 2 we explain an overview of
called a meiosis genetic algorithm (MGA) based on meiosis cell division following by showing how a meiosis
meiosis cell division. The meiosis cell division serves genetic algorithm works. Then, the third section will
producing chromosome diversity which speeds up illustrate our MGA. Experiments and their results are
converging to a mature solution chromosome. The given in section 4 and 5. The final section presents our
experiment results on a knapsack problem using the discussion and conclusion.
new algorithm show that it is more efficient than the
simple genetic algorithm. II. MEIOSIS CELL DIVISION

Duplication, mutation, and crossover in Meiosis Cell


Keywords-Meiosis Genetic Algorithm; Genetic Algorithm;
Knapsack Problem Division [6] are constructed for reproduction. MGA
applies these three strategies to produce better
chromosomes in all generation as follows:
I. INTRODUCTION
A. Duplication
Genetic Algorithms (GA) are search algorithms based A number of chromosomes are doubled by the
on natural selection and recombination mechanisms [1]. process of duplicating itself. There are two main types of
They combine the survival of the fittest and a randomized chromosome duplication: parity duplication and disparity
structured exchange of information in order to find exact duplication. In the first type of duplication, one child
or approximate solutions to optimization and search chromosome should be able to produce two twin-child
problems. In general, the mechanism of a simple GA chromosomes that inherit the same structure as their
requires five processes to come up with such an optimal parents. In the second type of duplication, one parent
solution: (1) problem representation by representing an chromosome is able to produce two children. However,
underlying problem with a suitable scheme; (2) the characteristics of one child will be the same as its
reproduction which is an artificial variant of the natural parent (leading strand), the other will not (lagging strand)
selection which follows the principle of Darwin, the [4]. In general, a mutation rate in lagging strand is higher
survival of the strongest creature; (3) crossover by than leading strand because, in lagging strand, they
randomly choosing two strings to be part of a new experience higher copy error rate at the point of joining
population ; (4) mutation in order to protect GA against as shown in Figure 1.
premature loss of information through crossover or
omitted reproduction; and (5) repeat the processes until
the gene-pool is empty or a stopping criterion is fulfilled.
However, GA has two major problems: a hill climbing
and a speed of optimal solution convergence. In order to
solve a speed problem, Maeda [3] applies the disparity
theory of an evolution and also proposes a new genetic
algorithm based on the disparity theory of an evolution
(DGA). Furthermore, to solve hill climbing problem,
Ronald [5] tries to reduce all redundant chromosomes by
checking repeated chromosome in both initial and
reproduction step. As a result, it brings chromosome
diversity and reduces hill climbing problems.
In this paper, we improve the simple GA performance Figure 1. Disparity Theory of Evolution.
by providing the algorithm with meiosis cell division [Source: http://nitro.biosci.arizona.edu/courses/EEB195-2007/Lectur
approaches, called “MGA”, for solving problems of hill e04/pics/image017.gif]

978-0-7695-3984-3/10 $26.00 © 2010 IEEE 285


DOI 10.1109/ITNG.2010.152
B. Mutation By utilizing mutation strategy, which affect only
This process generates different child chromosomes lagging strand, we use two points crossover in MGA and
than their parent chromosomes because of a copy error in select only strings with high fitness to our underlying
duplication step. problem and drop the rest. Finally, all processes are
repeated until the number of child chromosomes equal to
C. Crossover the population size.
Two chromosomes are combined with each other at
one point to exchange information at point of crossing. IV. EXPERIMENT STUDY

III. MEIOSIS GENETIC ALGORITHM Case 1. We conduct a crossover step only on lagging
strand using MGA; therefore, chromosome on leading
In this paper, we propose an MGA in order to generate strand will not change during crossover (see Figure 3).
chromosome diversity covering search space and inherit Case 2. We conduct a crossover step on leading strand
characteristics from its parent. MGA is a method to in MGA. As a result, the chromosome on leading strand
modify GA strategy by inserting a duplication step and will be changed during crossover step and become
assume that mutation step will start performing a diversify so it can cover most of our search (see Figure
crossover step later (see Figure 2). 4).
Case 3. We conduct a crossover step on lagging strand
in MGA and compare the results with a crossover on
leading strand in MGA.

Figure 3. Crossover on Lagging Strand.

Figure 4. Crossover on Leading Strand.

A. Performance Measurement
In order to analyze MGA performance, we use
maximum and average fitness, work speed, and evolution
speed as our indices. Work speed is the maximum of
Figure 2. Meiosis Genetic Algorithm.
MGA to GA ratio. Subsequently, evolution speed is the
maximum of child chromosome to parent chromosome
In MGA, we design an algorithm to produce child
ratio. Equations of work and speed are shown as follows:
chromosomes that inherit characteristics from their
parent. This step we called “a selection step”. We have
chosen our fitness function to make sure that this
operation will bring about only feasible strings. In
addition, we performed a chromosome duplication
process following the disparity theory of evolution [4].
Since parent chromosome will normally generate two
child chromosomes, leading strand and lagging strand, where Max is a maximum value of chromosome in
having these two types of child chromosomes will help us generation i. Work is a work speed for the first algorithm
find an optimal solution for problem(s) faster. compared with the second algorithm. The speed

286
measurement will present how GA and MGA evaluate
itself in order to provide an optimal solution.

B. Knapsack Problem
The name “Knapsack” arises from the rather unnatural
application of a hiker trying to fill her knapsack to
maximum total value. Each item she considers taking
with her has a certain value and a certain weight. An
overall weight limitation gives the single constraint.
Therefore, the standard knapsack problem considers a
number of items (n), each with a known value (v) and
weight (w), where v and w are independent. The goal is to
maximize the value obtained by selecting a subset of the
items whose collective weight does not exceed a given
capacity limit [7]. In certain contexts, the effective weight
of an item may itself be a decision variable.
Figure 5. A Comparison of Maximum and Average Fitness Value
However, problem occurs when we do not know how
between GA and MGA utilizing a Crossover on Lagging Strand.
many items we should select to knapsack in order to get
item’s maximum value and item’s weight summation. If Case 2. Similar to Case 1, we found that the maximum
the item weight summation is more than knapsack item and average fitness value of GA and MGA utilizing a
weight value, the item we choose will be set a weight crossover on leading strand are greater than GA (See
equal to zero due to the fact that we can not overload our Figure 6). In this case, the leading chromosome was
items. Therefore, the equation of knapsack problem is as evaluated by crossover, while the lagging chromosome
follows: was evaluated by mutation. Therefore, in MGA, every
child chromosome was evaluated in each generation and
be able to cover more search space.

where wj is a weight of item j; vj is a value of item j;


We is a upper bound of knapsack total weight; and if the
item j is selected to be in the knapsack, xj is equal to 1
and 0 otherwise.

C. Parameter Setting
In our experiment, a number of items in knapsack
problem are equal to 100 and we randomly assign value
and weight to items from [0,100]. Then, we assign weight
to knapsack by summarizing all item weights and divide
them by 2. For MGA and GA, we set a mutation rate in
leading strand equal to 0 and 0.2 in lagging strand
(mutation rate in GA is depending on lagging strand’s
mutation rate). In addition, we set a crossover rate equal
Figure 6. A Comparison of Maximum and Average Fitness Value
to 1 while a number of populations are 100. We run the between GA and MGA utilizing a Crossover on Leading Strand.
process in 300 generation and repeat the processes 10
times. In each round, we also randomly assign weight and Case 3. We observed MGA performance using
value to item in knapsack problem. crossover on leading strand and crossover on lagging
strand. We found that MGA with crossover on leading
V. EXPERIMENT RESULTS strand gives the maximum fitness better than MGA with
crossover on lagging strand because of chromosome
Case 1. After we conducted a crossover step on diversity.
lagging strand in MGA and compared the results with However, an average fitness value in MGA crossover
GA, we found that maximum and average fitness value of on leading strand can cover less search space than MGA
MGA is greater than GA because of chromosome crossover on lagging strand. This is because MGA
diversity. By this strategy, it can provide chromosomes crossover on lagging strand produces the best
with an ability to fast converge to optimal solution (see chromosomes among the ones obtained from crossover
Figure 5). on lagging strand and the good parent chromosomes.

287
a lot lower than MGA. Consequentially, the stability and
the speed in searching are lower. While MGA with the
crossover on lagging strand and crossover on leading
strand have a 0.0789% และ 0.0993% changing rate
respectively. As a result, MGA with the crossover on the
leading strand provide a faster search capacity than the
lagging strand.

Table II. AVERAGE EVOLUTION SPEED OF GA AND MGA

Genetic Algorithm 0.0422%


Meiosis GA Cx on Lagging Strand 0.0789%
Meiosis GA Cx on Leading Strand 0.0993%

Figure 7. Max and Average of MGA Crossover on Lagging Strand


compared with MGA Crossover on Leading Strand.

Figure 9. GA’s Evolution speed.

Figure 8. Work Speed of MGA Compared with GA.

Figure 8 shows the comparison between the work


speed of MGA crossover on leading strand and a
crossover on lagging strand. The average of the speed is
shown in Table I. We can see that MGA crossover on
leading strand provides an optimal solution faster than the
crossover on lagging strand because chromosome can
Figure 10. MGA crossover on lagging strand’s evolution speed.
cover search space better and can improve maximum
fitness more than parent prevent. Our result is similar to
Case 3.

TABLE I. A COMPARISON ON AVERAGE WORK SPEED OF


MGA and GA

Meiosis GA Cx on Lagging Strand 11.08%


Meiosis GA Cx on Leading Strand 17.31%

Evolution speed for GA, MGA crossover on leading


and lagging strand are shown in Figure 9, 10, and 11,
respectively and the average of speed is shown in Table
II.
Our finding indicates that the simple GA has only
0.0422 percent improvement in each generation, which is Figure 11. MGA crossover on leading strand’s evolution speed.

288
VI. CONCLUSION [9] Y.H. Song, G.S Wang, A.T. Johns and P.Y. Wang,
“Improved Genetic Algorithms with Fuzzy Logic Controlled
It is known that one serious disadvantage of Genetic Crossover and Mutation,” UKACC International Conference
algorithms in particular is that it is a time consuming on Control, 1996, 1, 2-5 Sept 1996, pp 140-144.
[10] R. Takahashi, “Solving the Traveling Salesman Problem
procedure. However, by applying meiosis cell division
through Genetic Algorithms with Changing Crossover
approach, it is possible to improve the efficiency of GA. Operators,” Fourth International Conference on Machine
In this paper a novel approach to GA efficiency Learning and Applications, 15-17 Dec. 2005, p. 6.
improvement has been presented. As we have shown, this
approach is capable of attaining a substantial
improvement of converging speed. In our experiment, we
conducted 2 sets of crossover strategies, a crossover on
lagging strand and a crossover on leading strand to
generate chromosome diversity. We found that the
evolution of chromosomes in MGA crossover on leading
and lagging strand is faster than GA. That is, MGA can
provide an optimal solution faster than GA.
Moreover, we compared parent preservation strategy
with chromosome diversity strategy. We found that the
later strategy can provide us with an optimal solution
faster than the former due to the reduction of
chromosome diversity by the former strategy in leading
strand. However, we perceived that the parent
preservation strategy gains more stability than the
diversity strategy.
Since both mutation and crossover in MGA are
random processes, it could lead to slow chromosome
evolution. Applying rough set theory [8] or fuzzy logic
[9] to calculate the point of crossover and mutation may
help us improve MGA performance. In addition, we
might be able to obtain fitting chromosomes faster if we
use advanced crossover operators [10].

REFERENCES

[1] J. Holland, “Adaptation in Natural and Artificial


Systems, ”The University of Michigan Genetic Algorithm
in Press, 1975.
[2] D. Charles, “On the Origin of Species by Means of Natural
Selection or the Preservation of Favored Races in the
Struggle for Life,” London, John Murray, 1859.
[3] Y. Maeda, “Modified Genetic Algorithms Based on
Disparity Theory of Evolution,” IFSA World Congress and
20th NAFIPS International Conference, July 2001 Joint 9th
Volume 4, pp. 2235-2240.
[4] M. Furusawa and H. Doi, “Promotion of Evolution:
Disparity in the Frequency of Strandspecific Misreading
Between the Lagging and Leading DNA Strands Enhances
Disproportionate Accumulation of Mutations,” J. theor.
Biol., vol. 157, pp. 127-133, 1992.
[5] S. Ronald, “Duplicate Genotypes in a Genetic Algorithm,”
IEEE International Conference on Evolutionary
Computation Proceedings, May 1998, pp. 793-798.
[6][online].Available: www.nfe.go.th/etv/document/know
ledge_lif/Bio11.pdf
[7] S. Martello and P. Toth, “Knapsack Problems: Algorithms
and Computer Implementations,” 1990, Wiley, New York.
[8] F. Li, Q.H. Liu, F. Min and G.W. Yang, “New Crossover
Operator based on the Rough Set Theory for Genetic
Algorithm,” Proceedings of 2005 International Conference
on Machine Learning and Cybernetics, 5, 18-21 Aug 2005,
pp. 2907-2912.

289

S-ar putea să vă placă și