Proc. of the Second Intl. Conf.

on Advances in Electronics, Electrical and Computer Engineering -- EEC 2013

Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-981-07-6935-2 doi:10.3850/ 978-981-07-6935-2_68

SOFT COMPUTING AND ITS COMPONENTS

MOHD TABISH KHAN

Abstract- Soft Computing plays a vital role in creation of

intelligent machines. Almost all the techniques that constitute
II. Artificial Neural Networks
Soft Computing are inspired by natural processes. The following
paper explores all such techniques and gives a nutshell of their A neuron is a cell in the brain whose principal
function is the collection, processing and dissemination of
working.
electrical signals. The brains information processing capacity
is thought to emerge primarily from networks of such neurons.
Keyword- Soft Computing, Neural Networks, Evolutionary
This is the main reason behind employing the artificial neural
Algorithm, Bayesian Network
networks for information processing.
Artificial Neural Network use learning algorithm by
I. Introduction our understanding of how the brain learns. It is a simplified
model of the biological neural system. It consists of highly
Soft Computing is a new multidisciplinary field that interconnected neural computing elements that have the ability
was proposed by Dr Lotfi A. Zadeh, whose goal was to to learn and thereby acquire knowledge and making it
construct a new generation Artificial Intelligence. This idea of available for use. Their practical application includes speech
Soft Computing was initiated in 1981 when Dr Zadeh recognition, objective recognition, image retrieval and the
published his first paper on Soft Data analysis. ability to recommend products that a user will like. Neural
Soft Computing is the fusion of methodologies that Network aims at finding similarity between human and
were designed to model and enable solutions to real world artificial information processing system.
problems, which are not modelled or too difficult to model, . An artificial neural network consists of an
mathematically. Soft Computing is applied to a field within interconnected group of artificial neurons called nodes
Computer Science which is characterised by the use of inexact working in a parallel way to solve the problem related to the
solution to computationally hard task such as the solution of aspect of classification. The nodes can be seen as
NP-Complete problems, for which there is no known computational units. They receive inputs, and process them to
algorithm that can compute an exact solution in polynomial obtain an output. This processing might be very simple (such
time. Unlike hard computing scheme, which strives for as summing the inputs), or quite complex (a node might
exactness and full truth, soft computing techniques exploit the contain another network...)
given tolerance of impression, partial truth and uncertainty for
a given problem. The connections determine the information flow
between nodes. They can be unidirectional, when the
Soft computing techniques resemble biological information flows only in one sense, and bidirectional, when
processes more closely than traditional techniques. The the information flows in either sense. The interactions of
learning ability of the human brain provides us with a new toll nodes though the connections lead to a global behaviour of the
for designing a cognitive machine that can learn and recognise network, which cannot be observed in the elements of the
complicated pattern like human faces, to understand distorted network. This global behaviour is said to be emergent. This
speech, decipher sloppy handwriting and comprehend nuance means that the abilities of the network supercede the ones of
of natural languages. It imitates the human mind to effectively its elements, making networks a very powerful tool.
employ modes of reasoning that are approximate rather than
exact. The most intriguing part of neural networks is its
paradigms of learning. Their types include supervised learning
The computing tools that are the major constituents in which it learns to predict an output when given an input
of soft computing includes neural networks, fuzzy systems, vector, Reinforcement learning in which it learns to select an
neuro-fuzzy systems, evolutionary algorithm, swarm action to maximise payoff and unsupervised learning under
intelligence and Bayesian Network. All these components will which it discovers a good internal representation of the input.
be discussed briefly in this paper.
The type of Neural Network includes Feed-forward
neural network and Recurrent Neural Network etc. Feed-
Mohd. Tabish Khan forward neural network are the commonest type of neural
network in practical application. It has a layered system in
Institute of Engineering and Technology, Indore which the first layer is the input and the last layer is the output
India and the remaining are the hidden layers. If there are more than
[email protected] one hidden layer then it is called deep neural network.

332
 Proc. of the Second Intl. Conf. on Advances in Electronics, Electrical and Computer Engineering -- EEC 2013
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-981-07-6935-2 doi:10.3850/ 978-981-07-6935-2_68

Recurrent Network have directed cycles in their connection. of the controller. This makes it easier to mechanize tasks that
These are relatively more natural ways to model sequential are already successfully performed by humans.
data. They are equivalent to very deep nets with one hidden
layer per time slice and also have the ability to remember IV. Neural Fuzzy System
information in their hidden state for a long time. But it is very
hard to train them to use this potential. Neural-fuzzy System refers to combinations
of artificial neural networks and fuzzy logic. Neural-fuzzy
system was proposed by J. S. R. Jang. Neural-fuzzy
Output Layer hybridization results in a hybrid intelligent system that
synergizes these two techniques by combining the human-like
reasoning style of fuzzy systems with the learning and
connectionist structure of neural networks. The reasons behind
combine these two paradigms is to overcome the difficulties
Hidden Layer and inherent limitations of each isolated paradigm. Neural
networks introduce its computational characteristics of
learning in the fuzzy systems and receive from them the
interpretation and clarity of systems representation.

Input Layer V. Evolutionary Computation

Evolutionary programming was introduced
Figure 1 Feed-forward Neural Network by Lawrence J. Fogel in the US. It is an area of computer
science that uses ideas from biological evolution to solve
computational problems in an optimized manner because
optimization is at the heart of many natural processes
including Darwinian evolution. Evolutionary computation uses
iterative progress, such as growth or development in a
population. This population is then selected in a
guided random search using parallel processing to achieve the
desired end.
Evolutionary computing techniques mostly
involve metaheuristic optimization algorithms. These include
fields like evolutionary algorithm, swarm intelligence. The
common underlying idea behind all these techniques is the
same, given a population of individuals the environmental
Figure 2 Recurrent Network with multiple hidden layers pressure causes natural selection (survival of the fittest) and
this causes a rise in the fitness of the population. Given a
quality function to be maximized we can randomly create a set
of candidate solution i.e., elements of the function’s domain,
and apply the quality function as an abstract fitness measure –
III. Fuzzy Systems the higher the better. Based on this fitness, some of the better
The fuzzy sets theory was initiated by Lofti A. Zadeh, candidates are chosen to seed the next generation by applying
in 1965. Fuzzy systems play a leading role in soft computing recombination and/or mutation to them. Recombination is an
and this stems from the fact that human reasoning is not crisp operator applied to two or more selected candidates (the
and admits degrees. The notion central to fuzzy systems is that children). Mutation is applied to one candidate and result is
truth values (in fuzzy logic) or membership values (in fuzzy one new candidate. Executing recombination and mutation
sets) are indicated by a value on the range [0, 1], with 0 leads to a set of new candidates (the off springs) that compete
representing absolute Falseness and 1 representing absolute – based on their fitness (and possibly age)- with the old ones
truth. Fuzzy systems are based on fuzzy logic (FL). Fuzzy for a place in the next generation. This process can be iterated
Logic is a form of many-valued logic or probabilistic logic; it until a candidate with sufficient quality (a solution) is found or
deals with reasoning that is approximate rather than fixed and a previously set computational limit is reached. A simple
exact. Fuzzy Logic incorporates a simple, rule-based IF X algorithm to define the functioning of evolutionary
AND Y THEN Z approach to a solving control problem rather computation can be given by:
than attempting to model a system mathematically. FL BEGIN
provides a simple way to arrive at a definite conclusion based
upon vague, ambiguous, imprecise, noisy, or missing input INITIALISE population with random candidates solution;
information. Fuzzy logic has the advantage that the solution to EVALUATE each candidate;
the problem can be cast in terms that human operators can
understand, so that their experience can be used in the design REPEAT until termination condition is satisfied
SELECT parents;

333
 Proc. of the Second Intl. Conf. on Advances in Electronics, Electrical and Computer Engineering -- EEC 2013
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-981-07-6935-2 doi:10.3850/ 978-981-07-6935-2_68

RECOMBINE pairs of parents; network represent a set of random variable nodes that take
discrete values. Two nodes should be connected directly
MUTATE the resulting offspring; because if one affects or causes changes to other, the arc
EVALUATE new candidates; indicates the direction of the effect. Also the values should be
both mutually exclusive and exhaustive,
SELECT individuals for the next generation;
A Bayesian Network has the following components
OD
 a set of nodes, one per variable
END
 a directed, acyclic graph (link = "directly influences")
A. Evolutionary Algorithm  a conditional distribution for each node given its
parents:
Evolutionary Algorithms are a kind of stochastic search
and optimization heuristics. Today these are successfully used Given by: P (Xi | Parents (Xi))
for solving numeric problems, such as optimization, automatic Applications of Bayesian network include decision
programming and so on. Evolutionary Algorithm have a making maximizing expected utility, adaptive testing. It may
conceptual base of simulating the evolution of individual also be used to update beliefs about states of certain variables
structure by well known processes such as selection, mutation when some other variables were observed, i.e., computing
and reproduction. Their importance is coming from their conditional probability distributions.
ability in optimizing complex function.
Conclusion
B. Swarm Intelligence
In this research paper an overview of the constituents
Swarm intelligence is the collective of soft computing is studied. The soft computing plays a very
behaviour of decentralized, self-organized systems, natural or significant role in machine learning. Thus for the process of
artificial. The concept is employed in work on artificial making the machines intelligent and inducing in them the
intelligence. The expression was introduced by Gerardo human like thinking characteristics, the soft computing could
Beni and Jing Wang in 1989, in the context of cellular robotic be very helpful.
systems. Swarm Intelligence (SI) is the property of a system
whereby the collective behaviours of (unsophisticated) agents Acknowledgment
interacting locally with their environment cause coherent
functional global patterns to emerge. SI provides a basis with The author would like to thank the professors of
which it is possible to explore collective (or distributed) Institute of Engineering and Technology, Indore for their
problem solving without centralized control or the provision of sincere guidance and helpful advice which provided a platform
a global model. It exploits the emergent collective intelligence to produce this paper.
of groups of simple agents.
This inspiration has comes from nature, especially References
biological systems like ant colonies, bee hives and so on The
Ants form and maintain a line to their food source by laying a [1] Zadeh, Lotfi A., "Fuzzy Logic, Neural Networks, and Soft
trail of pheromone, i.e. a chemical to which other members of Computing," Communication of the ACM, March 1994, Vol. 37 No. 3,
the same species are very sensitive. They deposit a certain pages 77-84.
amount of pheromone while walking, and each ant prefers to [2] McCulloch, Warren; Walter Pitts (1943). "A Logical Calculus of
follow a direction rich in pheromone. This enables the ant Ideas Immanent in Nervous Activity". Bulletin of Mathematical
colony to quickly find the shortest route. The first ants to Biophysics 5 (4): 115–133.
return should normally be those on the shortest route, so this [3] J. J. HOPFIELD Neural networks and physical systems with emergent
collective computational abilities. Proc. NatL Acad. Sci. USA Vol. 79,
will be the first to be doubly marked by pheromone (once in pp. 2554-2558, April 1982 Biophysics.
each direction). Thus other ants will be more attracted to this
[4] “Artificial Intelligence: A Modern Approach” 2nd Edition By Rusell and
route than to longer ones not yet doubly marked, which means Norvig Pg 719.
it will become even more strongly marked with pheromone. [5] Rumelhart, D.E; James McClelland (1986). Parallel Distributed
The main advantages of employing the swarm intelligence Processing: Explorations in the Microstructure of Cognition.
Cambridge: MIT Press.
includes performance optimization and robustness self-
[6] Novák, V., Perfilieva, I. and Močkoř, J. (1999) Mathematical principles
organized control and cooperation (decentralized) division of of fuzzy logic Dodrecht: Kluwer Academic.
labour and distributed task allocation indirect interactions. [7] Zadeh, L. A. et al. 1996 Fuzzy Sets, Fuzzy Logic, Fuzzy Systems, World
Scientific Press.
VI. Bayesian Network [8] R. Chiong, Th. Weise, Z. Michalewicz (Editors), Variants of
Evolutionary Algorithms for Real-World Applications, Springer, 2012.
The term "Bayesian networks" was coined by Judea [9] E. Eiben and M. Schoenauer, Evolutionary computing, Information
Pearl in 1985. Most commonly, BNs are considered to be Processing Letters, 82(1).
representations of joint probability distributions. A Bayesian [10] E. Goldberg. Genetic algorithms in search, optimization and machine
network is a graphical structure that allows us to represent and learning. Addison Wesley, 1989.
reason about an uncertain domain. The nodes in a Bayesian

334
 Proc. of the Second Intl. Conf. on Advances in Electronics, Electrical and Computer Engineering -- EEC 2013
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-981-07-6935-2 doi:10.3850/ 978-981-07-6935-2_68

[11] Beni, G., Wang, J. Swarm Intelligence in Cellular Robotic Systems,

Proceed. NATO Advanced Workshop on Robots and Biological
Systems, Tuscany, Italy, June 26–30 (1989).
[12] Ant Colony Optimization by Marco Dorigo and Thomas Stützle, MIT
Press, 2004.
[13] Eiben, A.E., Smith, J.E. (2003), Introduction to Evolutionary
Computing, Springer.
[14] J., Pearl (1994). "A Probabilistic Calculus of Actions". In Lopez de
Mantaras, R.; Poole, D.. UAI'94 Proceedings of the Tenth international
conference on Uncertainty in artificial intelligence. San Mateo CA:
Morgan Kaufman. pp. 454–462.

335