IRJET- The Essentials of Neural Networks and their Applications

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 371
The Essentials of Neural Networks and their Applications
Meghna Rawat1, Harsha Khandelwal2, Navneet Gupta3
1,2Student, Dept. of ECE, Arya Institute of Engineering & Technology, Jaipur, Rajasthan, India
3Associate Professor, Dept. of ECE, Arya Institute of Engineering & Technology, Jaipur, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract – This article looks at the essentials for artificial
intelligence and more specifically neural networking systems
in today's competitive business world. Some core principles of
neural network architecture are discussed, the advantages of
such networks. The domain of commercial applications of
neural technology has beenhighlighted. Neuralnetworkshave
various applications and the potential that exists in various
civil and military fields is tremendous.
Key Words: Artificial Neural Network (ANN), Neural
Network (NN), Artificial Intelligence,Feedback Network,
Neural Network Learning, Applications and drawbacks
of ANN.
1. INTRODUCTION
Technology has become very dynamic inthelastfewyears.It
is fuelling itself at an ever-increasing rate. Computers are a
major component of this entire revolution. Computers that
can help fight diseases by designing new drugs, computers
that can design better computers, computers that simulate
reality. This is a very exciting time for technology as
traditional boundaries are now becoming blurred.
We often think that computers can only decide Boolean
statements, whether that statement is true or false. Such
logical statements are joined together to form a series of
rules. To program a computer, all one has to do is define a
problem properly, write a specification, and use these rules.
The program tells the computer, from governance to
governance, exactly what to do. But it is difficult to program
a computer for more 'subjective' tasks, such as what is the
weather forecast or what will the price of gold be tomorrow.
These functions are impossible to define exactly. Complex
and incomplete patterns must be uniquelyidentified.Nature
is chaotic and we need something to understand this chaos.
Computers require a more 'human-like' ability, decision-
making ability, a different approach to guessing and
changing opinions. We humans learn by example and do not
need to look at every example to make a guess, judging by
what we have been taught.
As the emphasis is increasingly on autonomy, intelligence
and an increasing amount of information required by
businesses, traditional processing technology can only cope
with faster hardware with more complex bespoke software.
The growing problem in the 1990s and into the millennium
is that engineers no longer have the luxury of computing all
of those algorithms or identifying all the rules in these
complex systems. In fact, these systems are so complex and
chaotic and doing so would be a failure.
Given the high stakes and intense competition in all sectors
of the industry, intelligent business decisions are more
important than ever. There is an even more important case
for military applications. Data analysis plays an important
role in the trade and operations of the armed forces (both
peace-time and war-time)asimportantfactors. Theinherent
limitations of existing statistical technology make general
data analysis a very tedious and often costly process - it
requires assumptions, rigorous rules, data constraints as
well as extensive trial and error experiments and
programming. Interpreted errors, biases, and mistakes are
introduced. Valuable competitive insights are lost.
Technology based on artificial intelligence (AI) will soon
become the only way to create such a system economically.
2. NEURAL NETWORKS
Neuralcomputersarebasedonthebiologicalprocessesofthe
brain (human nervous system). Terms that can learn like the
brain, have been widely used in parallel, learning machines
and revolutionary to describe neural computing. It is not
surprising that most industries believe that taking a neural
approach would require specialized, expensive neural
integrated circuits, large parallel computers, or high-level
computers.
Traditional computers focus on emulating human thought
processes, rather than how they are actually received by the
human brain. However, neural computers adopt an
alternative way of directly demonstrating the biological
structure of the human brain and the way it processes
information (albeit at a much simpler level). This requires a
new type of architecture which, like the human brain,
consists of alarge number of heavily transferable processing
elements. Operating in parallel manner. Such architecture is
now technically and commerciallypossibletobedeployedon
a standard computer (from laptops and desktops to
mainframes) and is certain to increase in general use.
2.1 Basic Theory
Neural computing is a relatively new but rapidly expanding
branch of computing with its origins in the early 1940s.
Although this has been seen by traditional computing since
the 1960s, the field experienced fluctuations in popularity in
the late 1980s as a result of new developments and general
advances in computer hardware technology.

Neural networks are mathematical models, originally
inspired by biological processes in thehumanbrain.Theyare
constructed from a number of simple processing elements
that are connected by weighted pathways to form a network.
Each element calculates its outputas a non-linear functionof
its weighted inputs. When combined in a network, these
processing elements may be applied arbitrarily. Complex
non-linear tasks that can be used to solve classification,
prediction, or optimization problems.
Fig 1: Human NN vs. Artificial NN
Neural networks can be taught to performcomplextasksand
do not require programming as traditional computers. They
are largely parallel, extremely fast and intrinsically fault-
tolerant. They learn from experience, generalize from
examples, and are able to extract essential features from
noisy data. They require significantly shorter development
time and may respond to unspecified conditions or may not
be predicted earlier. They are ideally suited for real-world
applications and may provide solutions to currently
impossible or commercially impractical problems.
In simple terms, a neural network is made up of many
processing elementscalledneurons,whoseinterrelationsare
called synapse. Each neuron accepts input from the outside
world or from the outputofotherneurons.Theoutputsignals
from all neurons eventually propagate their influence over
the entire network to the last layer where the results can be
output to the real world. Synapse has a processing value or
weight, which is learned during the training of the network.
The functionality and power of the network mainly depends
on the number of neurons in the network, the
interconnectivity pattern or topology, and the value of the
load assigned for each synapse.
2.2Classification of Neural Networks
There are many artificial neural networks (ANNs). Just as
there are many ways to connect a circuittoperformaspecific
function with asinglecircuittopologyappliedtoallproblems,
the same is true for the mural network. The easiest to
understand and most commonly used architecture is a
globally connected feed-forward network -sometimescalled
multilayer assumption (MLP), typically trained with
backpropagation of error algorithms, From where vector
learn quantification,radialbasefunction,HopfieldandCohen,
etc.
Convolutional Neural Networks, or CNNs, were designed to
map image data to an output variable. They have proven so
effective that they are the go-to method for any type of
prediction problem involving image data as an input. The
benefit of using CNNs is their ability to develop an internal
representation of a two-dimensional image. This allows the
model to learn position and scale in variant structures in the
data, which is important when working with images.
Global interconnectivity means that all neuron outputs
(through their weight) of one layer are connected to the next
layer and to every neuron input. The inputs of neurons at the
input layer are from the outside world. Such networks
perform classification and optimizationoperationsverywell.
Neuron production values can be expressed mathematically,
but due to the underlying non-linear operators, these
equations provide intuitive insights into how neural
networks perform their functions.
1. Single-layer feedforward neural network
In a layered neural network, the neuronsareorganizedinthe
form of layers. The simplest structure is the single-layer
feedforward network that consists of input nodes connected
directly to the single layer of neurons. The node outputs are
based on the activation function.
2. Multilayer feedforward neural network
The second class of a feedforward neural network
distinguishes itself by the presence of one or more hidden
layers, whose computation nodes are correspondinglycalled
hidden neurons. By adding one or more hidden layers, the
network is enabled to extract higher-order statistics from its
input.
Fig 2: NN model parameters/networks
There are some additional features for this type of network
that usually apply to all neural networks regardless of
architecture. First, a neural network is over-specified,
meaning that there are many more unknowns than the
equations that describe the system. Secondly, there are
usually several weight sets (perhaps an infinitenumber)that
will solve the sameproblem.Finally,theweightsetoriginates
from training algorithms and is not programmed like
traditional algorithms. This training process relieves the

designer of developing an algorithm solution to the problem
at hand.
Some ANNs are classified as feed-forward, while others are
iterated (i.e., implement feedback) depending on how the
data is processed through the network. Another way to
classify ANN types is their learning method (or training), as
some ANNs employ supervised training, while others are
referred to as unprepared or self-organizing. Supervised
training corresponds to a student-directed instructor.
Clustering of data into similar clusters inevitably leads to an
unheard-of algorithm, which is based on measurement
characteristics or features serving as the input of the
algorithm. ANNs can be implemented in software or
specialized hardware.
2.3Neural Network Training
The process of calibrating the valuesofweightsandbiasesof
the network is called training of neural network to perform
the desired function correctly.
1. Supervised Learning
The basic aim is to approximate the mapping functionso
well that when there is a new input data (x) then the
corresponding output variable can be predicted. It is called
supervised learning because the process of an learning can
be thought of as a teacher who is supervising the entire
learning process. Thus, the “learning algorithm” iteratively
makes predictions on the training data and is corrected by
the “teacher”, and the learning stops when the algorithm
achieves an acceptable level of performance (or the desired
accuracy). In supervised learning, the data will be presented
in a form of couples (input, desired output), and then the
learning algorithm will adapt the weights and biases
depending on the error signal between the real output of
network and the desired output.
2. Unsupervised Learning
The main aim of Unsupervised learning is to model the
distribution in the data in order to learn more about the
data. It is called so, because there is no correct answer and
there is no such teacher. Algorithms are left to their own
devises to discover and present the interesting structure in
the data. In supervised learning,thedata will bepresented in
a form of couples (input, desired output), and then the
learning algorithm will adapt the weights and biases
depending on the error signal between the real output of
network and the desired output.
3. Reinforcement Learning
In reinforcement learning,anappropriateactionistaken
to maximize the reward in a particular situation. It is
employed by various software and machines so that it can
detect possible behavior or path in a specific situation.
Reinforcement learning differsfromsupervisedlearningina
way that supervised learning holds the answer key to
training data so the model is trained with thecorrectanswer
whereas in reinforcement learning, there is no answer but
the reinforcement agent decides that what to do in order to
complete the given task. In the absence of a training dataset,
it is bound to learn from its experience.
2.4 Advantages of Neural Networks
 Neural networks have the ability to self-learn and
produce that is not limited to the input provided to
them.
 Input is stored in its network instead of database;
Therefore, data loss does not affect its function.
 These networks can learn from examples and
implement them when a similar event occurs,
allowing them to work through real-time events.
 Even if a neuron is not responding or a piece of
information is missing, the network can detect a
fault and still produce an output.
 They are versatile and can performmultipletasksin
parallel without affecting the system performance.
2.5 Limitations of Neural Networks
 NN needs training to operate.
 The architecture of NN is different from the
architecture of microprocessors therefore need to
be emulated.
 Requires high processing time for large neural
network.
 NNs do not provide explanationsfortheirdecisions.
 NN decisions are not supported by significant tests,
hence low validity.
 High cost & complex network is another drawback
of NN.
4. APPLICATIONS OF NEURAL NETWORK
The Artificial Neural Network has been in existence since
1943, when it was initially designed, but has recently come
to light under Artificial Intelligence, due to applications that
make it much better. Artificial neural networkshavebecome
an accepted information analysis techniqueina widevariety
of disciplines. This has resulted in a wide variety of
commercial applications (both in product and service) of
neural network technology. Given below are the domains of
commercial applications of neural network technology:

1. Business
 Sales Forecasting
 Customer Research
 Digital Marketing
 Data Validation
 Real Estate
 Risk Management
2. Document & Form Processing
 Pre-processing
 Layout analysis
 Machine printed character recognition
 Character segmentation
 Graphics recognition
 Hand printed character recognition
 Signature verification
3. Finance Industry
 Market trading
 Fraud detection
 Credit rating
4. Food Industry
 Odour/aroma analysis
 Product development
 Quality assurance
5. Manufacturing Industry
 Process monitoring and control
 Quality control
 Identification
 Planning & scheduling
6. Science & Engineering
 Electrical engineering
 Agricultural Control System Engineering
 Civil engineering
7. Medical & Health Care Industry
 Image analysis
 Biochemical analysis
 Drug design
 Diagnostic system
8. Education
 OMR technology
 Estimatingstudentretention anddegreecompletion
time
 Data mining
9. Transportation & Communication
 Self-driving cars
 Automatic navigation system
 Indoor optical wireless communication
3. CONCLUSION
As we know that neural network is such a vast subject thatit
is not possible to cover completely in justa fewpages,yet we
try to give a glimpse of artificial neural network throughthis
paper. Here, we understood the fundamentals of neural
networks and their applications in various industries. It is
clear that the usage of neural network is going to increase in
the future in terms of both the number and type of their
applications. The future of neural networks is bright and
current research leads in the right direction towards the
ultimate goal of all artificial intelligence, namely, the
development of a humanoid robot that can work and think
like a human. However, we need to developmorealgorithms
and programs so that we can remove the limitations of
artificial neural networks and make it more useful fora wide
variety of applications. If the artificial neural network
concept is combined with computational automata, FPGA
and fuzzy logic, then we will certainly solve somelimitations
of neural network technology.
REFERENCES
[1] James A Anderson, An introduction to Neural Network.
Bradford Books
[2] Dan. W Patterson, Artificial Intelligence and Expert
Systems, PHI
[3] Kumar Satish, “Neural Networks” Tata Mc Graw Hill
[4] S. Rajsekaran & G.A. Vijayalakshmi Pai, “Neural
Networks, Fuzzy Logic and GeneticAlgorithm:Synthesis
and Applications” Prentice Hall of India.
[5] Siman Haykin,”Neural Networks” Prentice Hall of India
[6] Haykin S. Neural Networks and Learning Machines. 3rd
ed. Hamilton, Ontario, Canada: Pearson Education,
[7] Dhruba J Sharma. Susmita G Sarma, Research paper on
Neural networks and their applications in industry,
DESlDOC Bulletin of Information Technology.

[8] Widrow, Bernard & Lehr, Michael, A. Thirty Years of
adaptive neural networks: Perceptron, madaline and
backpropagation. (Proc. IEEE, September 1990).
[9] Amari, Shun-ichi. Mathematical foundations of
neurocomputing;(Proc.IEEE,September1990).Fausett,
L. (1994) Fundamentals of Neural Networks, Prentice
Hall, USA.
[10] Website reference for neural network learning,
‘www.geeksforgeeks.org’

IRJET- The Essentials of Neural Networks and their Applications

More Related Content

What's hot (20)

Similar to IRJET- The Essentials of Neural Networks and their Applications (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- The Essentials of Neural Networks and their Applications