Original Article International Journal of Research - GRANTHAALAYAH
ISSN (Online): 2350-0530 July 2021 9(7), 343–348
ISSN (Print): 2394-3629
MACHINE LEARNING: AN OVERVIEW
Adya Trisal1 and Dr. Dheeraj Mandloi2
1
Department of Computer Engineering, Institute of Engineering and Technology, Devi Ahilya University, Indore, India
2
Department of Applied Sciences, Institute of Engineering and Technology, Devi Ahilya University, Indore, India
ABSTRACT
Given the tremendous availability of data and computer power, there is a resur-
gence of interest in using data driven machine learning methods to solve issues
where traditional engineering solutions are hampered by modeling or algorith-
mic laws. The purpose of this article is to provide a comprehensive review
of machine learning, including its history, types, applications, limitations and
future prospects. In addition to this, the article also discusses the main point of
difference between the ield of arti icial intelligence and machine learning.
Keywords: Algorithms, Arti icial Intelligence, Machine learning, Reinforcement
Learning, Supervised Learning, Unsupervised Learning
Received 5 July 2021
Accepted 20 July 2021 1. INTRODUCTION
Published 31 July 2021
Machine learning (often abbreviated as ML) can be de ined as the study of computer
Corresponding Author
Dr. Dheeraj Mandloi, dmandloi@ie algorithms that ameliorate automatically through experience and by the advent of
[Link] data Mitchell (1997c). It is often seen as the subset of arti icial intelligence. A model
DOI 10.29121/ is built with the help of machine learning algorithms and is based on training data
granthaalayah.v9.i7.2021.4120 set also known as the sample data. The end goal of this model is to make predic-
Funding: This research received tions devoid of any external explicit programming. ML has its applications in almost
no speci ic grant from any funding
agency in the public, commercial, all the sectors. Medicine, email iltering, speech recognition, image recognition and
or not-for-pro it sectors. computer vision being a few. It is basically used in places where it is impractical to
Copyright: © 2021 The use conventional algorithms Hu et al. (2020). Machine learning is the process of com-
Author(s). This is an open access puters iguring out how to perform tasks without being speci ically trained to do so.
article distributed under the
terms of the Creative Commons It involves computers analyzing and learning trends from data. To perform simple
Attribution License, which tasks it is advisable to program algorithms which speci ically tell the computer how
permits unrestricted use,
distribution, and reproduction in to perform a task, in this process there is no learning done by the computer itself.
any medium, provided the original For more complex tasks, it can be strenuous for a human to manually generate the
author and source are credited.
required algorithms. In practicality, the approach of aiding the machine in develop-
ing its own logic and algorithm is more ef icacious than the traditional approach of
being programmed explicitly at every step Alpaydin (2020).
How to cite this article (APA): Trisal, A., & Mandloi, D. D. (2021). Machine learning: an overview. International Journal of Research 343
- GRANTHAALAYAH, 9(7), 343-348. doi: 10.29121/granthaalayah.v9.i7.2021.4120
Machine learning: an overview
2. MACHINE LEARNING THROUGH AGES
The expression, machine learning was devised by Arthur Samuel in the year of
1959 (an American innovator in the ield of computer gaming and arti icial intelli-
gence Samuel (1959) Kohavi and Provost (1998). In 1960s Nilsson in his book took
this idea further and toyed with the idea of pattern classi ication as an application
of ML Nilsson (1965). The idea of pattern recognition was so signi icant that it
continued inspiring the Ml pioneers in 1970s Duda and Hart (1973). Further
adding to these advancements, in the year 1981, a report was presented on using
teaching approaches to facilitate a neural network to recognize 40 characters
that included twenty-six letters, ten digits and four special symbols from a com-
puter terminal Bozinovski (1981). Later, Tom M. Mitchell stipulated a extensively
acknowledged de inition of the machine learning algorithms: ”A computer program
is said to learn from experience E with respect to some class of tasks T and per-
formance measure P if its performance at tasks in T, as measured by P, improves
with experience E.” Mitchell (1997a). The ‘tasks’ de ined in his de inition proposes
an operational de inition rather than elucidating the ield in cognitive and practical
terms. This shadows Alan Turing’s proposition in his paper ”Computing Machinery
and Intelligence” Harnad (2008). The present day ML has two primary purposes, the
irst one being to segregate data and the second one is to get the models to predict
the future results.
3. TYPES OF MACHINE LEARNING
There are various methods to train machine learning algorithms, each with their own
characteristic merits and demerits. Machine learning comprises of two variants of
data-labeled and unlabeled data. Labeled data comprises of both the input and out-
put parameters in a machine-understandable form. Unlabeled data on the other hand
has one or none of the parameters in a machine-understandable forma. All though
there are types of machine learning algorithms that are used in very speci ic use-
cases, but they can be broadly classi ied into 3 categories: supervised, unsupervised
and reinforcement learning.
Supervised Learning: Supervised learning algorithms shape a model of the given
dataset, which contains both the inputs, and the corresponding labeled outputs Rus-
sell and Norvig (2010). The dataset comprises of training examples that have input
mapped with the corresponding output. In the mathematical model, each training
example is characterized by an array known as the feature vector, and the training
data is embodied by a matrix Through iterative process supervised learning algo-
rithms formulates a function, which can be used to predict the output of the new
inputs Mohri et al. (2012). An algorithm that increases the accuracy of its predic-
tions gradually over time has learned to perform that task Mitchell (1997b). Super-
vised learning can be further bifurcated into active learning, classi ication and regres-
sion Alpaydin (2010). Classi ication algorithms are used when the outputs are a dis-
International Journal of Research - GRANTHAALAYAH
344
Trisal Adya and Mandloi Dr. Dheeraj
crete set of values, and regression algorithms are used when the outputs conforms
to a continuous range or function.
Unsupervised Learning: Unsupervised learning algorithms presume a set of
data that comprises nothing but the inputs, and ind a pattern in the data, for instance
grouping or clustering of data points. In this type of learning, the algorithms learn
from unlabeled test data. An Unsupervised learning algorithm does not respond to
feedback, rather it works by identifying commonalties. It is used for anomaly detec-
tion, clustering and inding the probability density function. Jordan et al. (2004)
Reinforcement Learning: Reinforcement learning is a ield of machine learning
involved with software agents and their action taking process pertaining an environ-
ment where the objective is to maximize the reward. Reinforcement learning forms
a part of many disciplines like game theory, simulation based intelligence, genetic
algorithms etc. Machine learning - Wikipedia (2021)
The environment is generally denoted as a Markov decision process. For rein-
forcement learning, dynamic programming is utilized Otterlo and Wiering (2012).
Reinforcement learning algorithms are used when exact models are not feasible.
These algorithms are used in automatic cars as well.
4. APPLICATIONS
Machine learning algorithms play a vital role in the conditions where development
is a necessity deployment. The ever-changing nature of machine learning solutions
is one of the core reasons for its global popularity. These algorithms are versatile in
nature and can be used as a substitute for some human activities. The best example
of this is the fact that customer services executives are being replaced by chatbots
which work by the advent of natural language processing. These chatbots work by
analyzing customer queries and provide support for the same without human inter-
vention. Large tech giants like Facebook, Net lix, Google, and Amazon deploy recom-
mendation systems that aids in providing exclusive and personalized content to indi-
vidual users based on their preferences. Facebook uses a system of recommendation
engines for advertisements and news feeds. Net lix accumulates data (information)
of its account holders and suggest variety of movies and shows based on the predilec-
tion of the user. Google on the other hand, uses machine learning to structurize its
results and is also used for YouTube’s recommendation system.
5. LIMITATIONS
Although machine learning has been transformative in some ields, machine-learning
programs often fail to deliver expected results ”Why Machine Learning Models Often
Fail to Learn: QuickTake Q&A”. [Link]. 2016-11-10. Archived from the
original on 2017-03-20. Retrieved 2017-04-10 (2016) “The First Wave of Corpo-
rate AI Is Doomed to Fail” (2018). There are various reasons for this: a lack of
International Journal of Research - GRANTHAALAYAH
345
Machine learning: an overview
(appropriate) data, an unavailability of data, data skew, con identiality issues, poorly
designed goals and algorithms, incorrect tools and personnel, insuf icient resources,
and evaluation problems ”9 Reasons why your machine learning project will fail”.
[Link]. Retrieved 2018-08-20 (2018). In 2018, an Uber automatic car
failed to identify a person, and the walker was killed as a result of the incident “Why
Uber’s self-driving car killed a pedestrian” (2018). Even after years of dedication and
investments of billions of dollars, IBM Watson’s endeavors to leverage machine learn-
ing in medicine were thwarted ”IBM’s Watson recommended ’unsafe and incorrect’
cancer treatments - STAT”. STAT. 2018-07-25. Retrieved 2018-08-21. (2018). Machine
learning has been used to update evidence in relation to systematic reviews and
increased reviewer workload as a direct outcome of biomedical literature. While
it has advanced with the aid of training sets, it has not yet progressed to the point
where it can minimize strain without compromising the required precision for the
discoveries.
6. AI VS ML
ML progresses to have applications in different ields as well as use-cases, it has
become more imperative to know how to distinguish between arti icial intelligence
and machine learning. The terminology ‘arti icial intelligence’ has been used as a gen-
eralized term to denote technology, which is capable of human like cognitive skills.
ML on the other hand, is a subset of Arti icial intelligence pertaining to algorithms
that can improve themselves without being explicitly programmed to do so. Machine
learning includes deep learning. Deep learning has neural networks as their process-
ing units, which are a type of algorithms that mimic neurons present in the human
brain.
7. FUTURE PROSPECTS
Machine Learning is based on the ideas of computer algorithms that learn through
trials and discoveries in a proactive manner. It is a form of Arti icial Intelligence
that allows program applications to predict outcomes with high precision. It dif-
ferentiates between building computer programs and aiding machines in memoriz-
ing without human involvement. Machine learning’s future looks extremely promis-
ing. Machine learning applications are being used in practically every mainstream
domain. Medicine, search engines, social media marketing, and academia, to name a
few, are all signi icant consumers of this technology. It appears effectively unattain-
able to ind a domain devoid of this technology. From a small enterprise to a MNC
the jobs which are presently being done manually shall be automated in the future.
Machine learning is known for almost every recent trend and pattern observed in the
literary circles, according to Gartner, the world’s largest research, advisory, and con-
sultatory institution. ML has the potential to revolutionize our lives in ways that were
International Journal of Research - GRANTHAALAYAH
346
Trisal Adya and Mandloi Dr. Dheeraj
previously unthinkable. Machine learning is known for every recent trend and pat-
tern seen in the literary circles, according to Gartner, the world’s largest research,
advisory, and consultatory institution, and rightfully so. Machine learning has the
potential to revolutionize our lives in ways that were previously unthinkable. Com-
putational rationalization and emerging Ml algorithms have reached a fundamental
turning point and will gradually develop and upgrade for all intents and purposes.
Developing advanced intelligent systems that learn, adapt, and even perhaps act on
their own rather than simply following predetermined instructions is a watershed
moment for innovators and technology providers. Individuals have sought to cre-
ate a machine that behaves and performs all activities in the same way that a human
does. As a consequence, Machine Learning has become AI’s ultimate blessing to the
mankind in terms of achieving the goals. Machine learning has continued to invade
new territories with unprecedented fervor. self-driving cars, computerized assis-
tants, robots, and green infrastructure have recently proved that smart machines
are achievable and could deliver appealing bene its. Numerous industrial sectors,
including commerce, manufacturing, construction, inance, health care, media, and
engineering, have been revolutionized by simulated intelligence modeled after the
human brain.
REFERENCES
”9 Reasons why your machine learning project will fail”. [Link]. Retrieved 2018-
08-20. (2018).
Alpaydin, E. (2010). Introduction to Machine Learning. MIT Press. p. 9.
Alpaydin, E. (2020). Introduction to Machine Learning (Fourth ed.). MIT. pp. xix, 1–3., 13–18.
Bozinovski, S. (1981). Teaching space: A representation concept for adaptive pattern classi i-
cation” COINS Technical Report No. 81-28, Computer and Information Science Depart-
ment, University of Massachusetts at Amherst, MA.
Duda, R., & Hart, P. (1973). Pattern Recognition and Scene Analysis, Wiley Interscience.
The First Wave of Corporate AI Is Doomed to Fail. (2018). Harvard Business Review.
Harnad, S. (2008). The Turing Test Sourcebook: Philosophical and Methodological Issues in
the Quest for the Thinking Computer, Kluwer. The Annotation Game: On Turing (1950)
on Computing, Machinery, and Intelligence, 23–66.
Hu, J., Niu, H., Carrasco, J., Lennox, B., & Arvin, F. (2020). Voronoi-Based Multi-Robot
Autonomous Exploration in Unknown Environments via Deep Reinforcement Learn-
ing. IEEE Transactions on Vehicular Technology.
”IBM’s Watson recommended ’unsafe and incorrect’ cancer treatments - STAT”. STAT. 2018-07-
25. Retrieved 2018-08-21. (2018).
Jordan, M. I., Bishop, & Christopher, M. (2004). ”Neural Networks”. In Allen B. Tucker (ed.). In
A. B. Tucker (Ed.), Computer Science Handbook, Second Edition (Section VII: Intelligent
Systems). Boca Raton, Florida: Chapman & Hall/CRC Press.
Kohavi, R., & Provost, F. (1998). Glossary of terms. Machine Learning, 30(2–3), 271–274.
Machine learning - Wikipedia. (2021).
Mitchell, T. (1997a). Machine Learning. McGraw Hill.
Mitchell, T. (1997b). Machine Learning. McGraw Hill. p. 2.
International Journal of Research - GRANTHAALAYAH
347
Machine learning: an overview
Mitchell, T. (1997c). Machine Learning. New York: McGraw Hill. ISBN 0-07-042807-7. OCLC
36417892..
Mohri, M., Rostamizadeh, A., Talwalkar, A., & Talwalkar. (2012). Foundations of Machine
Learning. The MIT Press.
Nilsson, N. (1965). Learning Machines, McGraw Hill.
Otterlo, M. V., & Wiering, M. (2012). Reinforcement learning and markov decision processes.
Reinforcement Learning. Adaptation, Learning, and Optimization, 12, 3–42. 10.1007/
978-3-642-27645-3_1
Russell, S. J., & Norvig, P. (2010). Arti icial Intelligence: A Modern Approach (Third ed.). Pren-
tice Hall. Arti icial Intelligence: A Modern Approach.
Samuel, A. (1959). Some Studies in Machine Learning Using the Game of Checkers, CiteSeerX
[Link].2254. IBM Journal of Research and Development, 3(3), 210–229.
”Why Machine Learning Models Often Fail to Learn: QuickTake Q&A”. [Link]. 2016-11-
10. Archived from the original on 2017-03-20. Retrieved 2017-04-10. (2016).
Why Uber’s self-driving car killed a pedestrian. (2018). The Economist.
International Journal of Research - GRANTHAALAYAH
348
