I.J. Modern Education and Computer Science, 2013, 12, 10-15
Published Online December 2013 in MECS (http://www.mecs-press.org/)
DOI: 10.5815/ijmecs.2013.12.02
Cuckoo Search Algorithm using Lèvy Flight: A
Review
Sangita Roy
ECE Department, Narula Institute of Technology, WBUT, Agarpara, K olkata, India.
roysangita@gmail.com
Sheli Sinha Chaudhuri
Electronics & Telecommunication Engineering Department, Jadavpur University, Kolkata, India.
shelism@rediffmail.com
Abstract ─ Cuckoo Search (CS) is a new met heuristic
algorithm. It is being used for solving optimization
problem. It was developed in 2009 by Xin- She Yang
and Susah Deb. Uniqueness of this algorithm is the
obligatory brood parasitism behavior of some cuckoo
species along with the Levy Flight behavior of some
birds and fruit flies. Cuckoo Hashing to Modified CS
have also been discussed in this paper. CS is also
validated using some test functions. After that CS
performance is compared with those of GAs and PSO. It
has been shown that CS is superior with respect to GAs
and PSO. At last, the effect of the experimental results
are discussed and proposed for future research.
Index terms ─ Cuckoo search, Levy Flight, Obligatory
brood parasitism, NP-hard problem, Markov Chain, Hill
climbing, Heavy-tailed algorithm.
I. INTRODUCTION
At present, problem solving optimization algorithms
are met heuristics and all of them are nature inspired. As
soon as they are emanated, they are accepted by
researchers and applied widely. Ant Algorithms were
inspired from the behavior of ants in the wild. Particle
Swarm Optimizations (PSO) was stimulated from the
world of fish and bird, whereas the Firefly Algorithm
was influenced by the flashing patterns of tropical
fireflies. All of these bio-inspired met heuristic
algorithms are being applied in problem solving
optimization such as NP-Hard problems: the Travelling
Sales Man Problem (TSP). The strength of all modern
met heuristic algorithms are the fact that they mimic the
best properties in nature , particularly biological systems
emerged from natural selection over millions of years ,
of them two important features can be simulated
numerically or algorithmically into two major features :
intensification, and diversification. Intensification tries
to search around the current best solutions and select the
best candidates or solutions, whereas diversification
utilizes the search space of the algorithm efficiently. CS
algorithm revolves around the behavior of obligatory
brood parasitism of some species of cuckoo as well as
Copyright © 2013 MECS
the Levy Flights (after the name of French
mathematician Paul Pierre Levy) of some birds and fruit
flies which follow the random walk of heavy tailed
probability distribution step size. These observations are
formulated into algorithms and implemented as a novel
and new idea. CS is compared with other already
existing popular optimization algorithms in connection
with numerous optimization problems.
Cuckoo Hashing is another computing technique for
resolving hash collisions of values in hash functions in a
table. It derives the concept from some cuckoo species
chicks pushing out of the nests other eggs or chicks at
the time of hatching. The idea was derived from Ramses
Pugh and Fleming Fiche Rodler in 2001. Since
hybridization is possible between the above mentioned
algorithms, as a result we can say that CS is derived
from Swarm Intelligence. By statistical analysis it has
been established that problem solving success of CS is
far superior to PSO (PSO2007) [1]. Modified CS (MCS)
is also made by Walton et al [2]. It delivers high
convergence rate which out performs other optimizers.
Basically it performs massively at high dimension bench
mark objective functions. Consequently it can be
applied on engineering problems. To be precise, the
modification stresses on the information exchange
between two best eggs or best solutions.
The paper is mainly organized according to the
paragraphs as: i) Study of cuckoo behavior and Levy
flight, ii) The primary concept behind the Cuckoo
search with pseudo code, iii) Realization and
mathematical results using different test functions, and
iv)Comparative study between PSO and GA. Finally in
conclusion it has been shown that less number of
parameters (two in the given study) of the CS algorithm
leads to its better efficiency. In future more complex
parametric study can be carried out with more promising
results in terms of the aim of speed of convergence. As a
result of this the computational cost reduces.
II. OBSERVATION ON CUCKOO BEHAVIOR AND
LÈVY FLIGHTS
A. Obligatory Brood Parasitism of Cucko
I.J. Modern Education and Computer Science, 2013, 12, 10-15
Cuckoo Search Algorithm using Lèvy Flight: A Review
The first impressions about cuckoos are their
beautiful tone. But apart from their sound, they have a
spectacular intrusion reproductive strategy. This is
known as brood parasitism. There are two types of
brood parasitism: non-obligatory and obligatory. In nonobligatory brood parasitism, cuckoo lays eggs in the
nest of conspecifics (i.e., same species) and in their own
nests as well. Examples: Bank Swallows, African
Weavers. In obligatory brood parasitism cuckoos lay
eggs in the nest of hetero-specifies, and do not require
building the nest of their own and incubate the eggs.
Examples: Brown- Headed Cowbirds and European
Cuckoos. It has been reported that 1% of all bird species
follow obligatory brood parasitism. Example: all
African Honey Guides, almost half of the species of
cuckoos, the Black Headed Duck in South America,
Shiny Cowbirds, Screaming Cowbirds, Bronze
Cowbirds and Giant Cowbirds. There are three types of
brood parasitism: intraspecific brood parasitism,
cooperative breeding, and nest takeover. Sometimes the
host birds engage in fight with intruder cuckoos. Either
the host birds may throw out the eggs if they find the
eggs of invaders or may leave the nest and rebuild the
nest elsewhere. The female cuckoos of the New World
brood parasitic Tapera who are expert to imitate the
colour and shape of the eggs of chosen host species, as a
result of which, the probability of their eggs to be
relinquished declines and intensifies their reproductively.
The time of egg-laying of some parasitic cuckoo is also
astonishing. Parasitic cuckoo lays eggs on the nest of
host birds. Host bird incubates both the eggs of its own
as well as the eggs of parasitic cuckoos. It is also
amazing that the parasitic eggs hatch slightly earlier
than that of the host eggs. Host birds incubate parasitic
chicks as their own. By natural instinct, cuckoo chicks
try to evict the host eggs randomly out of the nest for
the sake of its own existence and food. By investigation,
it has been found that cuckoo chicken imitate the sound
of host chicken to obtain more food from the host
mother.
B. Concept of Levy Flights
A Levy Flight can be thought of as a random walk
where the step size has a Levy tailed probability
distribution. The name Levy Flight came after the
French mathematician Paul Pierre Levy. The term Levy
Flight was coined by Benoit Mandelbrot who used
specific definition of the distribution of the step sizes.
Eventually Levy Flight term has been using to refer
discrete grid rather than continuous space. It is a
Markov Process. Exponential property of Levy Flight
gives it a scale invariant property and they are used to
model data for exhibiting/ showing clusters. In nature
many animals and insects follow the properties of Levy
F ligh t. Recen t s tud ies of Reyno ld s an d Fr ye
demonstrate that fruit flies or drosophila melanogaster
covers the skies by using numerous series of straight
flight paths/ routes followed by a sudden right angle
turn which is a Levy-flight-style intermittent scale free
Copyright © 2013 MECS
11
search pattern. Hunter-gatherer forage pattern exhibit
the typical feathers of Levy Flight, observed by
Ju/’bonsai on human behavior. Studies also show that
light rays follow Levy Flights in optical material [3].
Ultimately, it is being used in optimization search and
significant results are emerging.
III. THE BASIC IDEA BEHIND CUCKOO SEARCH
Cuckoo Search (CS) is a new met heuristic search
algorithm. It is characterized by three laws: each cuckoo
lays one egg at a time and disposes the egg at a
randomly chosen host nest, secondly the best nest with
excellent quality of eggs will carry over to the next
generation, and lastly the number of host nest is fixed
and the probability of parasitic egg-discovery by the
host bird is pa [0, 1]. The host bird can either throw
out the parasitic egg or forfeit that nest for a better new
nest. For practicality, in the later hypothesis it can be
assumed that fraction pa of the nests are substituted by
new nests (with new random solutions).In one form of
fitness the quality or fitness of a solution is proposed to
be proportional to the value of the objective function in
case of a maximization problem. Other form of fitness
can be proposed in the same manner as that of fitness
function of GAs. In a simple manner of representation,
each egg in a nest is represented by a solution and each
cuckoo egg symbolizes a new solution. The aim is to
exploit the new and probably finer/ superior solutions
(cuckoos) to substitute a not-so good solution in the
nests. More intricate cases like multiple eggs in a nest,
symbolizing a set of solution, can be incorporated in this
algorithm. Here the simplest approach, i.e., one egg in
one nest, has been applied. Guided by the above three
rules, the CS is abridged by the pseudo code:
Cuckoo Search via Le'vy Flights
begin
Objective function f(x), x=(x1, x2,………xd )r
Generate initial population of
A host nests xi (i=1, 2, 3……, n)
While (l < Maxgeneration) or (stop criterion )
Get a cuckoo randomly by Levy Flights
Evaluate its quality/fitness Fi
Choose a nest among n (say, j) randomly
if (Fi > F j),
Replace j by the new solution;
end
A function (pa) of worse nests
Are abandoned and new ones are built;
Keep the best solutions
(or nests with quality solutions);
Rank the solutions and find the current best
end while
Postprocess results and visualization
end
I.J. Modern Education and Computer Science, 2013, 12, 10-15
12
Cuckoo Search Algorithm using Lèvy Flight: A Review
A new solution is symbolized by x (t+1) and a cuckoo
is represented by i, then a Levy Flight is carried out by
Where α > 0 is the step size which is adjusted
according to the scale of the problem of interests. In
most of the cases α is assumed to be unity. Equation 1
represents the stochastic equation of random walk.
Generally random walk is charecterised by Markov
chain where the next state is represented by the present
location (the first term of the above equation) and the
transition probability (the second term). EX-OR
symbolizes exclusive OR i.e., entry wise multiplication.
This entry wise multiplication is equivalent as that of
PSO, except that the random walk via Levy Flight is
more efficient in exploring / acquiring the search space
because its much longer step size increases
exponentially in the last stage. Levy Flight is a random
walk with the random step size following a Levy
distribution.
makes CS adaptable to the class of Meta population
algorithms.
IV. REALISATION AND NUMERICAL
VERIFICATION
A. Authentication and Specification Studies
After implementation, the algorithm has to be verified
using test functions with analytical or known solutions.
In this case, the bi-variate Michaelwicz function is used.
Where m=10 and (x, y) Є [0, 5] × [0, 5] with global
minima f * ≈ڏ-1.8013 at (2.20319, 1.57049).
Levy distribution has infinite variance with infinite
mean with a power-law step size of a heavy tail as it is
shown in fig 1.
Figure 2. The landscape of bi-variate Michaelwicz Function [5]
Figure 1. Levy and Normal Distribution [4]
Using Levy walk, few new solutions must be
explored for the speed up of local search from the best
solutions so far. It should be kept in mind as a warning
that the system must not be trapped in a local optimum.
For this reason, a portion or adequate amount of fraction
of the new solutions must be cropped up from far field
randomization with locations far enough from the
current best solution. At a glance, both CS and hillclimbing in combination with few large scale
randomizations appeared to be similar, but there is some
salient dissimilarity. Firstly, like GAs and PSO, CS is a
population based algorithm, at the same time it explores
some kind of elitism and/or selection as that of harmony
search. Secondly, the randomization is more efficient as
the step size is heavy tailed with any possible large step
size. Thirdly, the number of tuning parameters is less
than GAs and PSO which in turn promises to be
acceptable to a wider class of optimization problems.
Moreover, an individual nest representing a solution
Copyright © 2013 MECS
The landscape of this function is shown in fig 2. CS
finds these global optima easily and it is shown in fig. 3.
With final locations of the nests marked ◊. Here, the
number of nests, step size and probability of parasitic
egg-discovery used are n=15, α = 1 and pa = 0.25. In
most of the simulation cases n=15 to 50. The results
from the figure show that most of the nests accumulate
towards the global optimum as the optimum is
approaching. In case of multimodal functions, the nests
are scattered at different local optima. From the above
discussion, it is clear if the numbers of nests are higher
than that of the number of local optima, the CS scans all
the optima simultaneously. In case of multimodal and
multiobjective optimization problems, this advantage
may be more pronounced. In this particular case, the
number of host nests (i.e., the population size n) and the
probability
pa
are
varied
where
n=5,10,15,20,50,100,150,250,500 and
pa=
0,0.01,0.05,0.1,0.15,0.2,0.25,0.4,0.5.Simulation result
shows that n=15 and pa =0.25 are adequate for majority
of the optimization problems. From the results and
analysis, it has been deduced that the rate of
convergence mostly is independent of the parameters
used, which in turn leads to inessentiality of fine
adjustment for any problems. Therefore, throughout the
given problem, n=15 and pa = 0.25 has been used.
I.J. Modern Education and Computer Science, 2013, 12, 10-15
Cuckoo Search Algorithm using Lèvy Flight: A Review
13
Where a single global minima f* =0 at x* =(0,0,-----,0)
for all -600 ≤ xi ≤ 600 where i=1,2,---,d.
5) Ackley’s Function is multimodal
√
[
Figure 3. Search paths of nests in Cuckoo Search where the
final location of the nests are denoted by ◊ in the figure [6]
B. Test Functions
There are numerous benchmark test functions to
verify the performance of optimization algorithms. All
new optimization algorithms must be validated and
tested against these benchmark functions. In this
algorithm (CS), few test functions have been utilized for
simulations.
∑
[ ∑
]
]
Where a global minima f* = 0 at x* = (0,0,---,0)
between -32.768 ≤ xi ≤ 32.768 with i=1,2,----,d.
6) Generalized Rosen rock’s Function
∑[
]
1) Sphere Function-De Jong’s first function
Where a minima f(x*) = 0 at x* = (1, 1, ---, 1).
[
∑
]
7) Sahwefel’s Test Function is multimodal
Where global minimum f* =0at x* = (0, 0, -----, 0) with d
dimension.
2) Easom’s Test Function is unimodal
[
]
–
√| |
∑
Where a global minima f* = -418.9829d at xi* = 420.9687
with i=1, 2, ----d.
8) Rastrigin’s Test Function
∑
Where
[
] [
]
With global minimum of f* = -1 at (Π, Π) in a very small
region.
3) Shubert’s Bivariate Function
∑
]
[
]∑
4) Griewangh’s Test Function
This function has many local minimas.
–∏
Copyright © 2013 MECS
(
√
9) Michalewicz’s Test Function
It has d! local optima
[
(
)]
With 0 ≤ xi ≤ Π and i=1, 2, ---, d. The global minima is
f* ≈ -1.801 for d=2, whereas f* ≈ -4.6877 for d=5.
Where 18 global minimas exists in the region (x,y) [10,10] x [-10,10]. The global minimas have a value of
-186.7309.
∑
]
Where a global minima f* =0 at x* =(0,0,---0) in a
hypercube -5.12 ≤ xi ≤ 5.12 with i=1,2,---,d.
∑
[
[
)
C. Comparative Study of CS with PSO and GA
According to the recent research, PSO algorithms left
behind GAs and other traditional algorithms for many
optimization problems. This is because of the attributes
in the broadcasting ability of the current best estimates
which promise better and quicker convergence of
optimality. It is here been shown the performance of CS
versus that of PSO and GAs using most of the standard
test functions. The algorithms were run more than
hundred times for implementation of Matlab simulation
to get meaningful statistical interpretation. The
algorithms stop for the variation function values which
I.J. Modern Education and Computer Science, 2013, 12, 10-15
14
Cuckoo Search Algorithm using Lèvy Flight: A Review
are less than threshold tolerance ε ≤ 10-5. The Table 1
shows the results when the global optimas are reached.
The format of the table is: average number of
evaluations (success rate), i.e., 3221± 519(100%)
indicate the average number (mean) of the function
evaluation is 3221 with standard deviation 519. The
success rate of searching for the global optima for this
algorithm is 100%. It is also clear that CS is far more
efficient in searching the global optima with the highest
(100%) success rate. These evaluation processes of each
test function are practically instantaneous with a 3 GHz
desktop (personal) computer performing 10,000
evaluations in 5 seconds. For all the aforesaid test
functions, CS has outperformed both GAs and PSO. The
main reasons are: excellent balance between
randomisation and intensification with few number of
control parameters.
An efficient metaheuristic algorithm is specified by a
good balance between intensive local search strategy
and an efficient exploration of the whole search space.
Moreover, there are only two search parameters in this
algorithm; the population size n and pa. Once n is fixed,
pa primarily supervi\ses the elitism and in turn controls
between the randomization and local search. Due to
involvement of two parameters the CS algorithm
becomes less complex and hence more fundamental.
After original CS, there are few new researches that
have come out with more complex parameters taking
into account. More marvelous results are generated
ending up with the aim of speed of convergence.
TABLE I. COMPARISON OF CS WITH GENETIC ALGORITHMS
Algor
ithms
/Func
tionss
GA
PSO
CS
Michalew
iez’s
(d=16)
Rosenbr
ock’s
(d=16)
De
Jong’s
(d=256)
Schwefe
l’s
(d=128)
Ackley
’s
(d=128
)
Rastrigi
n’s
Easom’s
Griewan
k’s
Shubert
’s (18
minima)
52124±
3277
(98%)
89325±
7914
(95%)
55732±
8901
(90%)
25412±
1237
(100%)
227329
±7572
(95%)
32720±
3327
(90%)
110523
±5199
(77%)
19239±
3307
(92%)
70925±
7652
(90%)
54077±
4997
(89%)
3719±
205
(97%)
6922±
537
(98%)
32756±
5325
(98%)
17040±
1123(10
0%)
14522±
1275
(97%)
23407±
4325
(92%)
79491±
3715
(90%)
17273±
2929
(90%)
55970±
4223
(92%)
23992±
37557
(92%)
3221±
519
(100%)
5923±
1937
(100%)
4971±
754
(100%)
8829±
625
(100%)
4936±
903
(100%)
10354±
3755
(100%)
6751±
1902
(100%)
10912±
4050
(100%)
9770±
3592
(100%)
Multiple
Peaks
927±
105
(100%)
V. CONCLUSION
Authors have studied the CS algorithm which is
mainly centered on some species of adult cuckoos’
astonishing obligatory brood parasitism along with Levy
Flight behavior. Yang and Deb opened up a new avenue
by introducing the
vy-flight-style rather than the
simple random walk. The proposed CS algorithm is
validated and compared with some classical Algorithms:
GAs and PSO. From simulation and comparison, it is
evident that CS is much better and efficient than the
GAs and PSO for multimodal objective functions. This
is because of the fact that fewer parameters are to be
fine controlled in case of CS. It is also clear from the
earlier discussion that once n, the number of nests, is
fixed, only one parameter pa is left to be tuned. It should
also be noted that the convergence rate in CS using all
the aforesaid test functions are very fast and efficient
without the interference of pa. That indicates that these
two parameters do not require any fine control for a
problem of interest. Consequently CS is more generic
and robust for many optimization problems in
comparison with other met heuristic algorithms. This
promising robust optimization algorithm can effortlessly
be adapted to investigate multi-objective optimization
Copyright © 2013 MECS
approach with various constraints, even to NP-hard
problem. More investigations can be carried out with
more complex parameters for sensitivity of the
optimization and derive any possible relationship out of
the problem and the speed of convergence as well as the
computational cost. Although CS is a subfield of Swarm
Intelligence based algorithms, hybridization can be
developed with PSO, GAs and other algorithms which
will overcome the drawbacks of CS and will produce
robust outcome.
REFERENCES
[1] Civicioglu. P., Besdok. E., A conceptual
comparison of Cuckoo-Search, particle search
optimisation, differential evolution and artificial
bee colony algorithms, Springer Science and
Business Media B. V.2011.
[2] Walton S., Hassan O., Morgan K., and Brown M.
R., Modified cuckoo search: A new gradient free
optimisation algorithm, Chois, Solutions and
Fractals, 44, issue 9, pp 710-718(September 2011).
[3] Barthemy P., Bertolotti J., Wiersma D. S., A Levy
Flight for light, Nature, 453, 495-498(2008).
I.J. Modern Education and Computer Science, 2013, 12, 10-15
Cuckoo Search Algorithm using Lèvy Flight: A Review
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
http://www.google.co.in/imgres?q=levy+distributi
on&hl=en&sa=X&tbo=d&biw=1360&bih=629&t
bm=isch&tbnid=S2sLMB8a2_KFsM:&imgrefurl=
http://toshiclark.xanga.com/726841489/affirmative
-action- redux/&docid=v5KrG_1jRGKsiM&
imgurl=http://i43.tinypic.com/2216o0.png&w=496
&h=393&ei=zcKpUMupMcSmrAehsoDoAw&zoo
m=1&iact=rc&dur=327&sig=1023991376222725
30481&page=1&tbnh=138&tbnw=164&start=0&n
dsp=20&ved=1t:429,r:2,s:0,i:77&tx=110&ty=68.
X.-S. Yan, Harmony Search as a Metaheuristic
Algorithm, in Music-Inspired Harmony Search
Algorithm: Theory and Applications, Studies in
Computational Intelligence,Springer Berlin, vol.
191,pp. 1-44(2009).
http://www.metaheuristic.com/x_algorithm_metah
euristic_optimization.php.
Bonabeau E., Dorigo M., Theraulaz G., Swarm
Intelligance: From Natural to Artificial Systems.
Oxford University Press, (1999).
Bulm C. and Roli A., Metaheuristics in
combinatorial optimization: Overview and
conceptural comparison, ACM Comput. Surv,
35,268-308(2003).
Barthelemy P., Bertelotti J., Wiersma D. S., A
Lévy flight for light.Nature, 453, 495-498(2008).
Goldberg D. E., Genetic Algorithms in Search,
Optimisation and Mechine Learning, Reading.
Mass: Addision Wesley (1989).
Kennedy J. and Eberhart R. C.: Particle swarm
optimization.Proc.
of
IEEE
International
Conference on Neural Networks. Piscataway, NJ.
pp. 1942-1948 (1995).
Yang X. S., Nature-Inspired Metaheuristic
Algorithms, Luniver Press, (2008).
Yang X. S., Biology-derived algorithms in
engineering optimization (Chapter 32), in
Handbook of Bioinspired Algorithms and
Applications (eds Olarius & Zomaya), Chapman &
Hall/ CRC (2005).
Shlesinger M. F., Zaslavsky G. M. and Frisch U.
(Eds), Lévy Flights and Related Topics in Physics,
Springer, (1995).
Shilane D., Martikainen J., Dudoit S., Ovaska S. J.,
A general framework for statistical performance
comparison
of
evolutionary
computation
algorithms, Informational Sciences: an Int. Journal,
178,2870-2879(2008).
Shang Y. W., Qiu Y. H., A note on the extended
rosenrbock function, Evolutionary Computation,
14,119-126(2006).
Schoen F., A wide class of test functions for global
optimization, J. Global Optimization, 3,133-137,
(1993).
Kennedy J., Eberhart R., Shi Y.: Swarm
intelligence, Academic Press, (2001).
Payne R. B., Sorenson M. D., and Kiltz K., The
Cuckoos, Oxford University Press, (2005).
Reynolds A. M. and Frye M. A. ,Free-flight odor
tracking in Drosophilia is consistent with an
Copyright © 2013 MECS
15
optimal intermittent scale—free search, PLoS
One,2,e354(2007).
[20] Xin-She Yang,Suash Deb,Cuckoo Search via
e’vy Flights. Proc. Of World Congress on Nature
& Biologically Inspired Computing (NaBIC 2009),
India. IEEE Publications, USA.
[21] Shlesinger M. F, Search Research, Nature,
443,281- 282(2006).
Sangita Roy is an Assistant Professor at ECE
Department of Narula Institute of Technology under
WBUT .She has a teaching experience of more than
sixteen years. She was in instrumentation industry for
two years and in administration for two years. She
completed her Diploma (ETCE), A.M.I.E (ECE) and MTech (Comm. Engg).Currently perusing her PhD under
Dr. Sheli Sinha Chaudhuri at ETCE Department of
Jadavpur University. She is a member of IEI, IETE,
FOSET, and IEEE.
Dr. Sheli Sinha Chaudhuri is an Associate Professor at
ETCE Department of Jaduvpur University. She
completed her B-Tech, M-Tech, and PhD at Jadavpur
University. She has a vast teaching experience of 14
years. She has large number of papers in International
and national level journals as well as conferences.
Currently research scholars are pursuing PhD under her
guidance. She is a member of IEEE and IEI.
I.J. Modern Education and Computer Science, 2013, 12, 10-15