Cuckoo Search: An Introduction

Xin-She Yang

Middlesex University London

For details, please read my book:

Nature-Inspired Optimization Algorithms, Elsevier, (2014).

Matlab codes are downloadable from

https://uk.mathworks.com/matlabcentral/profile/authors/3659939-xs-yang

Xin-She Yang Book/Slides 1 / 16

 Cuckoo Search Almost Everything is Optimization

Almost Everything is Optimization

Almost everything is optimization ... or needs optimization ...
Maximize efficiency, accuracy, profit, performance, sustainability, ...
Minimize costs, wastage, energy consumption, travel distance/time,
CO2 emission, impact on environment, ...

Mathematical Optimization
Objectives: maximize or minimize f (x) = [f1 (x), f2 (x), ..., fm (x)],

x = (x1 , x2 , ..., xD ) ∈ RD ,

subject to multiple equality and/or inequality design constraints:

hi (x) = 0, (i = 1, 2, ..., M ),

gj (x) ≤ 0, (j = 1, 2, ..., N ).
In case of m = 1, it becomes a single-objective optimization problem.

Xin-She Yang Book/Slides 2 / 16

 Cuckoo Search

Optimization problems can usually be very difficult to solve, especially large-scale,

nonlinear, multimodal problems.
In general, we can solve only 3 types of optimization problems:
Linear programming
Convex optimization
Problems that can be converted into the above two
Everything else seems difficult, especially for large-scale problems.
For example, combinatorial problems tend to be really hard – NP-hard!
Deep Learning
The objective in deep nets may be convex, but the domain is not convex and it’s a
high-dimensional problem.
n i2
1 Xh
Minimize E(w) = ui (xi , w) − ȳi ,
n i=1

subject to various constraints.

Xin-She Yang Book/Slides 3 / 16

 Cuckoo Search Key Components for Optimization

Key Components for Optimization

Algorithm

Select Handle
Optimization
Solutions Constraints

Evaluate
Objectives

Xin-She Yang Book/Slides 4 / 16

 Optimization Techniques Optimization Techniques

Optimization Techniques
There are a wide spectrum of optimization techniques and tools.
Traditional techniques
Linear programming (LP) and mixed integer programming.
Convex optimization and quadratic programming.
Nonlinear programming: Newton’s method, trust-region method, interior
point method, ..., barrier Method, ... etc.
But most real-world problems are not linear or convex, thus traditional techniques
often struggle to cope, or simply do not work...

New Trends – Nature-Inspired Metaheuristic Approaches

Evolutionary algorithms (evolutionary strategy, genetic algorithms)
Swarm intelligence (e.g., ant colony optimization, particle swarm
optimization, firefly algorithm, cuckoo search, ...)
Stochastic, population-based, nature-inspired optimization algorithms

Xin-She Yang Book/Slides 5 / 16

 The Essence of an Algorithm The Essence of an Algorithm

The Essence of an Algorithm

Essence of an Optimization Algorithm
To generate a better solution point x(t+1) (a solution vector) from an existing
solution x(t) . That is, x(t+1) = A(x(t) , α) where α is a set of parameters.

x(1)
x(t)
?

x(0) x(t+1)

Population-based algorithms use multiple, interacting paths.

Different algorithms
Different ways for generating new solutions!

Xin-She Yang Book/Slides 6 / 16

 The Essence of an Algorithm Main Problems with Traditional Algorithms

Main Problems with Traditional Algorithms

What’s Wrong with Traditional Algorithms?
Traditional algorithms are mostly local search, thus they cannot guarantee
global optimality (except for linear and convex optimization).
Results often depend on the initial starting points (except linear and convex
problems). Methods tend to be problem-specific (e.g., k-opt, branch and
bound).
Struggle to cope problems with discontinuity.

Nature-Inspired Optimization Algorithms

Heuristic or metaheuristic algorithms (e.g., ant colony optimization, particle
swarm optimization, firefly algorithm, bat algorithm, cuckoo search, differential
evolution, flower pollination algorithm, etc) tend to be a global optimizer so as to
Increase the probability of finding the global optimality (as a global optimizer)
Solve a wider class of problems (treating them as a black-box)
Draw inspiration from nature (e.g., swarm intelligence)
But they can be potentially more computationally expensive.
Xin-She Yang Book/Slides 7 / 16
 The Essence of an Algorithm Cuckoo Search

Cuckoo Search
Cuckoo search (CS) was developed by Xin-She Yang and Suash Deb in 2009.

Cuckoo brood parasitism

59 cuckoo species (among 141 cuckoo species) engage the so-called obligate
reproduction parasitism strategy.
Cuckoos lay eggs in the nests of host birds (such as warblers) and let host birds
raise their chicks.
Eggs may be discovered/abandoned with a probability (pa ≈ 0.25).
Co-evolutionary arms race between cuckoo species and host species.

Cuckoo Behaviour (BBC Video)

Xin-She Yang Book/Slides 8 / 16
 The Essence of an Algorithm

Cuckoos’ Behaviour and Idealization (Yang and Deb, 2009)

Each cuckoo lays one egg at a time and dumps it in a randomly chosen nest.
The best nests with high-quality eggs will be carried over to the next generations.
The number of available host nests is fixed, and the egg laid by a cuckoo is
discovered by the host bird with a probability pa ∈ (0, 1). In this case, the host bird
can either get rid of the egg or simply abandon the nest and build a completely new
nest elsewhere.

Search Domain

Nest (Host)

Lévy flight
Nest (New location)
xt+1
i = xti + α L(s, λ).

Each egg/solution is replaced with a probability of pa

Here, xi is the solution vector (or position of nest i) in the search space at iteration t,
and α is a scaling factor. L(s, λ) is the step size to be drawn from the Lévy distribution
with an exponent λ.
Xin-She Yang Book/Slides 9 / 16
 The Essence of an Algorithm Cuckoo Search (CS) (Yang and Deb, 2009)

Cuckoo Search (CS) (Yang and Deb, 2009)

Two search mechanisms in CS: local random walks and global Lévy flights.
Local random walks:
xt+1
i = xti + s ⊗ H(pa − ǫ) ⊗ (xtj − xtk ).
[xi , xj , xk are 3 different solutions, H(u) is a Heaviside function, ǫ is a random
number drawn from a uniform distribution, and s is the step size.

Global random walks via Lévy flights:

λΓ(λ) sin(πλ/2) 1
xt+1
i = xti + αL(s, λ), L(s, λ) ∼ , (s ≫ s0 ).
π s1+λ

Generation of new moves by Lévy flights, random walks and elitism.

The switch between these two search mechanisms is governed by the discovery
probability pa = 0.25.

Xin-She Yang Book/Slides 10 / 16

 The Essence of an Algorithm Mathematical Foundation for Cuckoo Search

Mathematical Foundation for Cuckoo Search

Isotropic andom walks (diffusion) Lévy flights (superdiffusion)
Gaussian distribution Lévy distribution
1 h (s − µ)2 i Z
1 ∞ λ
p(s) = √ exp − 2
, L(s, λ) ∼ cos(ts)e−αt dt.
σ 2π 2σ π 0

Typical paths of t = 50 consecutive steps of random walks

√
Diffusion distance: d(t) ∼ t d(t) ∼ t(3−λ)/2 (for 1 ≤ λ ≤ 2)

Xin-She Yang Book/Slides 11 / 16

 The Essence of an Algorithm

Typical Parameter Values

Population size: n = 10 to 40 (up to 100 if necessary).
Lévy exponent: λ = 1.5.
α = O(L/100) to O(L/10) where L is the typical scale of the problem. Typically,
we can use α = 0.01 to 0.1 for function optimization.
Number of iterations tmax = 500 to 1000.

Pseudo-random step size (s) for Lévy flights

Quite tricky to generate, though Mantegna’s algorithm works well.
U
s= , U ∼ N (0, σ 2 ), V ∼ N (0, 1),
|V |1/λ

where ‘∼’ means ‘to draw’ random numbers from the probability distribution on the
right-hand side. The variance σ 2 is calculated by
h Γ(1 + λ) sin(πλ/2) i1/λ
σ2 = · ,
Γ((1 + λ)/2) λ2(λ−1)/2

where Γ(ν) is the standard Gamma function. For example, if λ = 1, we have σ 2 = 1

since Γ(1 + λ) = 1, Γ((1 + λ)/2) = 1 and sin(π/2) = 1.
Xin-She Yang Book/Slides 12 / 16
 The Essence of an Algorithm Cuckoo Search Pseudocode

Cuckoo Search Pseudocode

Algorithm 1: Cuckoo Search
Data: Objective functions f (x)
Result: Best or optimal solution
1 Initialization of parameters (n, pa , λ and α);
2 Generate initial population of n host nests xi ;
3 while (t <MaxGeneration) or (stop criterion) do
4 Get a cuckoo randomly;
5 Generate a solution by Lévy flights;
6 Evaluate its solution quality or objective value fi ;
7 Choose a nest among n (say, j) randomly;
8 if (fi < fj ) then
9 Replace j by the new solution i;
10 end
11 A fraction (pa ) of worse nests are abandoned;
12 New nests/solutions are built/generated;
13 Keep best solutions (or nests with quality solutions);
14 Rank the solutions and find the current best solution;
15 Update t ← t + 1;
16 end
Xin-She Yang Book/Slides 13 / 16
 The Essence of an Algorithm CS is very efficient

CS is very efficient
Cuckoo Search Demo: Highly Efficient!

Rosenbrock (banana) function

f (x, y) = (1 − x)2 + 100(y − x2 )2 , (x, y) ∈ R2 .

ln[1+f (x,y)]
=⇒

Cuckoo Search (Demo video at Youtube) [Please click to start]

Xin-She Yang Book/Slides 14 / 16

 The Essence of an Algorithm Multi-objective Cuckoo Search (MOCS)

Multi-objective Cuckoo Search (MOCS)

For example, the so-called ZDT function with D = 30 dimensions
minimize f1 (x) = x1 , and f2 (x) = g(x)h(x), x ∈ [0, 1]30 ,
where s
D=30
9 X f1 f1
g(x) = 1 + xj , h(x) = 1 − − sin(10πf1 ),
29 j=2 g g
has a nonconvex Pareto front in the domain 0 ≤ xi ≤ 1 where i = 1, 2, ..., 30.

Cuckoo Search (Demo video at Youtube) [Please click to start]

Xin-She Yang Book/Slides 15 / 16
 The Essence of an Algorithm Cuckoo Search (Demo Codes) and References

Cuckoo Search (Demo Codes) and References

CS Demo Codes
The standard CS demo in Matlab can be found at the Mathswork File Exchange
https://uk.mathworks.com/matlabcentral/fileexchange/74767-the-standard-cuckoo-search-cs

The multi-objective cuckoo search (MOCS) code is also available at

https://uk.mathworks.com/matlabcentral/fileexchange/74752-multiobjective-cuckoo-search-mocs

Some References
Xin-She Yang and Suash Deb, Cuckoo search via Lévy flights, In: Proceedings of the World Congress on
Nature & Biologically Inspired Computing (NaBIC 2009), IEEE Publications, pp.210-214 (2009).
Xin-She Yang and Suash Deb, Engineering optimisation by cuckoo search, Int. J. Mathematical
Modelling and Numerical Optimisation, vol. 1, no. 4, 330–343 (2010).
Xin-She Yang and Suash Deb, Multiobjective cuckoo search for design optimization, Computers &
Operations Research, vol. 40, no. 6, 1616–1624 (2013).
Xin-She Yang and Suash Deb, Cuckoo search: recent advances and applications, Neural Computing and
Applications, vol. 24, no. 1, 169–174 (2014).
Xin-She Yang, Cuckoo Search and Firefly Algorithm: Theory and Applications, Springer, (2013).
Xin-She Yang, Nature-Inspired Optimization Algorithms, Elsevier Insights, (2014).

Xin-She Yang Book/Slides 16 / 16