A Comparative study of Nature-Inspired

Optimization Approaches
Priya Arora* V.K Panchal**
[email protected] [email protected]

*
Department of CSE, Baddi University of Emerging Sciences & Technology, Himachal Pradesh.
**
Defence Terrain Research Laboratory, DRDO, Delhi.

Abstract heads, by seeking inspiration from nature i.e.

imitating the way nature works. Natural computing
A vast variety of population-based has taught us to think naturally about computation
optimization techniques have been and also to think computationally about nature [1].
formulated in recent years for use in different Our aim is to discuss why Nature has been such an
appealing choice for dealing with conventionally
engineering applications, most of which are
hard to solve problems. Conventional computing
inspired by natural processes taking place in paradigms often have difficulty dealing with real
our environment.This paper deals with the world problems, such as those characterized by noisy
comparative study of research work done in or incomplete data or multimodality, because of their
the field of Swarm Intelligence. Nature inflexible construction [8]. Natural systems have
Inspired Paradigms play a vital role in evolved over millennia to solve such problems, and,
solving the problems as they provide more when closely examined, these systems often contain
significant and attractive solutions. Nature many simple elements that, when working together,
inspired techniques are adopted because they produce complex emergent behavior [1]. Nature
inspired approaches have applied to a diverse set of
are adaptable, evolvable, resilient and easy to
problems because it requires an overview and not an
implement. Unlike conventional workgroups extensive description of the problem. Most of the real
that they have standard operating procedures world problems have components of both
to react to predetermined stimuli, nature imprecision and bias, and the general rule is that the
inspired paradigms can adjust to new greater the imprecision component (relative to the
situations and evolves quickly. The main bias component), greater the potential for Nature
objective of this paper is to provide an Inspired solution [5]. They have inspired several
overview of various nature inspired natural computing paradigms that can be used where
optimization and explore its various conventional computing techniques perform
techniques unsatisfactorily. One reason of that is because those
most traditional approaches are deterministic and
nature inspired is, in its majority, composed of
stochastic methods [8].
Keywords: - Optimization, Nature - inspired
optimization paradigms. This paper is organized into four sections. The
section two, following introduction describes what is
optimization and a broad classification of
1. Introduction optimization methods. The third section introduces
Nature inspired paradigms have evolved from the the concept of swarm intelligence and various
era of Alan Turin- Von Neumann architecture for representatives of swarm intelligence. Next section
solving complicated problems to the stage of describes the behavior of nature inspired techniques.
Fifth section presents concluding remarks of our
Computational Biology [5]. These days, not only
discussion.
computer scientists and engineers, but also business
analysts are developing solutions to their hardest
problems without scratching and cracking their
2. What is Optimization? Optimization methods

Optimization has been an active research field for

several decades. It is indeed a part of our everyday
life. Optimization is a mathematical discipline that
concerns the finding of minima and maxima of Deterministic Stochastic
functions, subject to so-called constraints.
Optimization originated in the 1940s, when George
Dantzig used mathematical techniques for generating Grid Search Clustering
"programs" (training timetables and schedules) for
military application. Since then, his "linear
programming" techniques and their descendents were
Covering methods Random Search
applied to a wide variety of problems, from the
scheduling of production facilities, to yield
management in airlines. Today, optimization Figure 2 Classification of optimization
comprises a wide variety of techniques from
Operations Research, artificial intelligence and Deterministic approaches include grid search,
computer science. Optimization, in general, is covering methods. Grid search does not exploit
concerned with finding an alternative or a best information of previous optimization steps, but rather
solutions for a given problem with the most cost assesses the quality of points lying on a grid over the
effective or highest achievable performance under the search space. Obviously, the grid density plays a
given the constraints or several criteria that are crucial role on the final points lying on a grid over
usually problem or user dependent [17]. Constraints the search space. On the other hand, covering
can be posed either by the user or problem itself, methods aim at the detection and exclusion of parts of
therefore reducing the number of solutions. If a the search space that do not contain the global
solution fulfils all the constraints it is called a feasible minimizer. All these mentioned approaches have well
or appropriate solution. Among all feasible solutions, studied theoretical backgrounds since their operation
Global Optimization problem concerns with the is based on strong mathematical assumptions [17].
detection of optimal solution. However, this is not Stochastic methods include random search,
always possible, although there may be some clustering methods. Clusterings strived to address a
suboptimal solutions which are acceptable, crucial deficiency of the generalized techniques.
depending on their quality or appropriability Moreover, various iterative schemes may converge
compared to the optimal one; this is described as on the same problem. This effect adds significant
local optimization [17]. An optimization algorithm computational cost to the algorithm and reduces its
used search for an optimized solution by iteratively performance. Random search methods are based on
transforming a current candidate solution into a probability distributions to produce samples of search
hopefully better solution. points [17]. Obviously output accuracy increases as
sample size approaches to infinity. In practice, this
2.1 Classification of Optimization methods approach is inefficient since it requires a vast number
of function evaluations to produce acceptable results.
Generally two major categories of optimization Significant efforts towards the improvement of
(figure 2) can be distinguished Deterministic and random search methods have lead to the development
Stochastic as proposed by Archetti & Schoen, 1984; of novel algorithms that draw inspiration from
Dixon & Szego, in 1978[17] . stochastic phenomena in nature [17]. These
approaches are called heuristics or met heuristics and
Deterministic approaches are characterized by the they simulate fundamental elements and procedures
exact reproducibility of the steps taken by the that produce evolution and intelligent behaviors in
algorithm, in the same problem and initial conditions. natural systems. Natural intelligence computational
Stochastic approaches produce samples of paradigms (NICP) is a incorporation of two
prospective solutions in the search space iteratively. stipulations natural intelligence and computational
paradigms which denotes that natural intelligence
can be computed under umbrella of computational
paradigms[20](figure 3).
 today, this family comprises many well - known
Natural Intelligence computational Paradigms (NICP)
Optimization techniques such as Biogeography
Based Optimization (BBO), Ant colony Optimization
(ACO), Particle Swarm Optimization (PSO), Plate
Tectonics. Etc.
Bio-EmergedConcoct by human mind
Nature as computing
Substrate 3.1 Biogeography Based Optimization
Artificial Fuzzy System DNA Computing Biogeography is the study of the
Neural
geographical distribution of biological
networks
organisms [7]. Mathematical equations
Rough Sets
Quantum that govern the distribution of organisms
Evolutionary Computing Computing were first discovered and developed
during the 1960s. The mindset of the
engineer is that we can learn from nature.
Granular computing
Artificial Immune System This motivates the application of
biogeography to optimization problems.
Just as the mathematics of biological
Perception based omputing genetics inspired the development of
Membrane Computing
genetic algorithms (GAs) and the
mathematics of biological neurons inspired
Anticipatory Computing the development of artificial neural
Amorphous Computing networks, mathematics of biogeography
as the basis for the development of a new
WisTech Computing
field: biogeography-based optimization
(BBO). Mathematical models of biogeography
Swarm Intelligence
describe how species migrate from one island to
another, how new species arise, and how species
become extinct [19]. The species are migrated to the
Figure 3 Showing Envisaged Taxonomy suitable habitats i.e., feature islands. These habitats
are actually the decisions, for the resultant groups.
Various methods decide which species are moved to
In recent years, metaheuristic, a family of stochastic which habitat using a fitness function [12]. The
techniques has become an active research area i.e. a fitness function is actually the information that is
computational paradigm straddling at times shared among all the habitats in order to decide the
evolutionary computation and neural computation is suitable habitat for migration of each species.
swarm intelligence[9].

3. Swarm Intelligence 3.2 Ant Colony Optimization

Dorigo (1992) introduced a stochastic optimization

The term swarm intelligence (SI) was coined in the
algorithm for combinatorial problems. The algorithm
late 1980s to refer to cellular robotic systems in
was inspired by the behaviour of ants in search of
which a collection of simple agents in an
food, and was named ant colony optimization. Ant
environment interact according to local rules [11].
colony optimization algorithms are inspired by the
The main inspiration behind the development of SI
fact that ants are able to find the shortest route
stems directly from nature. Fish schools, bird flocks,
between their nest and a food source, even though
ant colonies and animal herds, with their amazing
they are almost blind. This is accomplished by using
self- organization capabilities and reactions, produce
pheromone chemical! Trails as a form of indirect
collective behaviors that cannot be described simply
communication. Ants deposit pheromone trails
by aggregating the behavior of each team member. It
whenever they travel. The path taken by individual
is a property of systems of unintelligent agents of
ants from the nest in search for a food source is
limited individual capabilities exhibiting collectively
essentially random [3]. However, when many ants
intelligent behavior [18]. SI can be defined as higher
are searching for a food source simultaneously, the
- level frameworks aimed at efficiently and
paths taken are affected by the pheromone trails laid
effectively exploring a search space. Subsequently, a
by other ants. When ants encounter pheromone trails,
lot of diverse methods have been proposed, and
there is a higher probability that trails with higher intelligent technique for developing a computational
pheromone intensities will be chosen as more ants model. The major assumption is that it is the entropy
travel on paths with higher pheromone intensities, the which is the driving force leading to the formation of
pheromone on these paths builds up further, making it heterogeneous regions called as plates, similar to the
more likely to be chosen by other ants. The way this convection force in the mantle of the Earth.
form of positive reinforcement can be used to find the This technique is an evolution of a new paradigm, a
shortest path between the nest and a food source [5]. never before concept, the development of a new
algorithm based on Plate tectonics The plate tectonics
3.3. Particle Swarm Optimization based optimization algorithm can be used for the
purpose of water body extraction in satellite images
Particle Swarm Optimization (PSO) is a population in future developments for the extraction of more
based stochastic optimization technique inspired by land cover features from any multi-spectral satellite
social behavior of bird flocking. PSO applies the image. This algorithm is a major development in the
concept of social interaction to problem solving. It field of Artificial Intelligence& Image Classification.
was developed in 1995 by James Kennedy and Russ Also, this algorithm produces comparable results
Eberhart [Kennedy, J. and Eberhart, R. (1995). [13] with the other established AI techniques such as the
[11]. One of the important factors that particle swarm BBO [12].
optimization is attractive is simple that very few
parameters are there to adjust. It can achieve optimal
or near optimal solution in rather a short time. The 4. Behavior of Nature Inspired Techniques
particle swarm algorithm samples the search-space
by modifying the velocity of each particle. The When Complex natural phenomena are analyzed in
inspiration taken from the social-psychological terms of computational problems, our understanding
sciences suggests that individuals (particles) should of both nature and essence of computation is
be influenced by their own previous experience and enhanced. The main idea of this branch is to develop
the experience of its neighbors. Neighborhood here computational tools i.e. algorithms by taking
refers to topological similarity in a given structure of inspiration from nature for the solution of complex
the population [4]. There are a number of different problems. Characteristic for human designed
schemes to connect the individuals of the population. computing inspired by nature is the metaphorical use
Most particle swarm implementations use one of two of concepts, principles and mechanisms underlying
simple sociometric principles. The first, called gbest natural systems.
(g for global), conceptually connects all members of
the population to one another. The effect of this is i. Emergent Behavior: - Emergent behavior
that each particle is influenced by the very best is essentially any behavior of a system that
performance of any member of the entire population. is not a property of any of the components
The second, termed lbest (l for local), creates a of that system. This is the behavior that is
neighborhood for each individual comprising itself not attributed to any individual agent, but is
and its k-nearest neighbors in the population [13]. a global outcome of the agent group. A
collection of intelligent agents can produce a
better solution to a problem than the sum of
3.4 Plate Tectonics
the abilities of all agents when they work
individually [15].
Recent advances in remote sensing have widened
the platform for research in science and technology. ii. Self Similarity: - Nature provides
Undoubtedly the estimation of geo-physical examples of self-similarity for instance
properties of the land cover features like water, ferns, coastlines, mountains, cauliflowers,
urban, vegetation, rocky and barren areas play an our lungs, our brains and our kidneys.
important role in environmental, transportation and Indeed Nature is all about symmetric things
region planning, natural disaster, industrial and which differentiate it from the Euclidean
agricultural production[12]. Till now, natural geometry. The concept of self-similarity
computation and bio-inspired intelligent techniques carries through for a number of dissections.
like DNA computing, membrane computing, genetic As a result this symmetry can lead to the
algorithms, neural computing have been used for development of various structures which
demonstrating the applications of computational would be able to solve complex structures.
intelligence in the field of remote sensing. However,
geo-science has never been used as a nature inspired
 iii. Adaptability: - Swarm intelligent systems their flexibility and robustness. However, nature
involve functional interactions with the inspired optimization methods are not only tools that
environment. Living organisms can respond only neither provide solutions to every problems, nor
to or anticipate changes in the environment. are they always the most suitable and efficient
Hence, they can deal with dynamic approaches. These approaches are applicable to those
environment through self adaption [2]. problems which are complex i.e. involves a large
number of variables, highly dynamic, nonlinear etc.
iv. Ability to Evolve: - It is not a property of a Even in the situation where a single solution is not
single individual, but of a population of good enough or when diversity is important.
individuals.
References:
v. Flexible :- The flexibility of swarm
intelligent system to change quickly in a 1. De Casro N., editor. Fundamentals of natural
dynamic environment leads to its computing: basic concepts, algorithms, and
applications, 2006.
applicability to different problems
2. Bck, T., Fogel, D., Michalewicz, Z., editors.
Handbook of Evolutionary Computation. IOP
vi. Robust: - Swarms can deal with dynamic Publishing, U.K., 1997.
environment through self adaption which 3. Bonabeau E, Dorigo M, Thraulaz G., editors.
leads an aspect of the developmental pattern Swarm intelligence: from natural to artificial
of the organism which enables or enhances systems. Oxford University Press; 1999.
the probability of surviving. 4. Konstantinos E. Parsopoulos & Michael N.
Vrahatis, editors. Particle swarm Optimization
vii. Autonomous: - The swarm intelligent and Intelligence, Advances and applications,
2010.
system can function without human
5. Dorigo M., Optimization, Learning and Natural
intervention. It is usually understood as the Algorithms, PhD thesis, Politecnico di Milano,
process by which systems of many Italie, 1992.
components tend to reach a particular state, 6. Tomas Piatrik and Ebroul Izquierdo , Image
with no external inference. Classification Using an Ant Colony
Optimization Approach, LNCS 4306, pp. 159
viii. Decentralized: - The characterizing 168, 2008
property of a swarm intelligence system is 7. Dan Simon editor, Biogeography Based
its ability to act in a coordinated way Optimization, IEEE Transactions on
without the presence of a coordinator or of Evolutionary Computation, Vol. 12, No. 6, 2008
8. Lila Kari., Grzegorz Rozenberg., editors. " The
an external controller, without a central
Many Facets of Natural Computing,"
authority. Communication of the ACM,vol. 51,
no.10,October2008.
ix. Applicability to non - linear simultaneous 9. A. Engelbrecht. editor. Fundamentals of
problems: - It is very important to use Computational Swarm Intelligence. Wiley and
swarm intelligence to understand how large Sons, 2005.
scale emergent patterns and behaviors can 10. G Beni., J Wang., editors. Swarm intelligence.
result from the actions and interactions of In: Proc of the 7th annual meeting of the
the individual components of a system. robotics society of Japan. 1989. p. 4258.
11. White T, Pagurek B. editors, towards multi-
swarm problem solving in networks. In: Proc of
5. Conclusions the 3rd Int Conf on multi-agent systems
(ICMAS98).1998. p. 33340.
Natural processes provide great inspiration for 12. Goel Lavika., Gupta Daya., Panchal V.K.,
various problems that can be solved by modeling the "Biogeography and Plate Tectonics based
natural phenomena. We led to a deeper and broader Optimization for Water body Extraction in
understanding of the nature inspired optimization Satellite Images, International Conference on
approaches and how these approaches are forming a Soft Computing and Pattern Recognition
bridge to the standard methods of optimization. (SocProS), Advances in Soft Computing series
Natural computing has taught us that various natural of Springer Publications, Vol. 131, pp. 1-10,
2011.
phenomena based models may be used as basis for
13. Particle Swarm Optimization, Proceedings of
the development of novel optimization techniques for the 1995 IEEE International Conference on
various complex problems. The advantages of these Neural Networks, pp. 1942-1948, IEEE Press.]
approaches over the traditional approaches are due to
14. Andries P. Engelbrech.t, editor, Computational 18. Christian Blum Daniel Merkle, editors, Swarm
Intelligence, an Introduction. Second Edition, Intelligence, Introduction and Applications.
2010. Springer, 2008.
15. Julian Togelius., Optimization, Imitation and 19. Panchal, V.K.; Bhugra, D.; Goel, S.; Singhania,
Innovation: Computational Intelligence and V., editors, Decision gap within Swarm in
Games, a Thesis submitted for the degree of Natural terrain and feauture Extraction
Doctor of Philosophy Department of Computer Electronics Computer Technology (ICECT), 3rd
Science University of Essex September 2007. International Conference, 2011.
16. Leandro Minku., Introduction to Natural 20. Sana Nabi., Ashraf Shahid., waani A.M.,
Computation Lecture 1 Module Introduction Panchal V.K., editors. Exploring the Regime of
and Examples1 1Notes adapted from Intro to Natural Computing, 6th International
NC Year 2010. Multiconference on Intelligent Systems,
17. Yoel Tenne, Chi-Keong Goh., editors, Sustainable, New and Renewable Energy
Computational Intelligence in Optimization, Technology & Nanotechnology March,
Applications and Implementation. Springer,
2012.
2010.