Вы находитесь на странице: 1из 6

Machine Learning And its Approaches

Pagadala Vamsi Krishna, 171FA04227

Manoj Kumar Reddy, 171FA04511
CSE Department, Vignan’s Foundation for Science Technology and Research

Abstract –

With the growing engrossment of everyone on the uses of Computer and the
Internet, the world is advancing very much faster every minute. The usage of data is
also increasing second by second as large number of people are relying on the internet
for each and every aspect of their life. The storage and transmission rate of the data has
also increased drastically. Computer with more efficient processors that can handle all
the tasks have come into existence. With all these there is certainly a necessity of
analysing of data for extracting useful information and to develop a procedure based on
that analysis. The machine(computer) learns from the feedback and does not make a
mistake the next time. Machine Learning is basically a branch of artificial
intelligence that gives systems the ability to learn automatically and improve
themselves from the experience without being explicitly programmed or without the
intervention of human. Its main aim is to make computers learn automatically from the

I. Introduction networks", these were

mostly perceptron and other models that
were later found to be reinventions of
The name machine learning was coined the generalized linear models of
in 1959 by Arthur Samuel. Arthur Samuel, statistics. Probabilistic reasoning was also
an American pioneer in the field employed, especially in automated medical
of computer gaming and artificial diagnosis.
intelligence, coined the term "Machine
However, an increasing emphasis on
Learning" in 1959 while at IBM. As a
the logical, knowledge-based
scientific endeavour, machine learning
approach caused a rift between AI and
grew out of the quest for artificial
machine learning. Probabilistic systems
intelligence. Already in the early days of AI
were plagued by theoretical and practical
as an academic discipline, some researchers
problems of data acquisition and
were interested in having machines learn
representation. By 1980, expert
from data. They attempted to approach the
systems had come to dominate AI, and
problem with various symbolic methods, as
statistics was out of favour. Work on
well as what were then termed "neural
symbolic/knowledge-based learning did that supervised learning algorithm analyses
continue within AI, leading to inductive the training data (set of training examples)
logic programming, but the more statistical and produces a correct outcome from
line of research was now outside the field of labelled data.
AI proper, in pattern
recognition and information retrieval.
training of machine using information that
Machine learning, reorganized as a separate is neither classified nor labelled and
field, started to flourish in the 1990s. The allowing the algorithm to act on that
field changed its goal from achieving information without guidance. Here the
artificial intelligence to tackling solvable task of machine is to group unsorted
information according to similarities,
problems of a practical nature. It shifted
patterns and differences without any prior
focus away from the symbolic approaches training of data. Unlike supervised learning,
it had inherited from AI, and toward NO LABEL is provided that means no
methods and models borrowed from training will be given to the machine.
statistics and probability theory. It also Therefore, machine is restricted to find the
benefited from the increasing availability of hidden structure in unlabelled data by our-
digitized information, and the ability to self.
distribute it via the Internet.
working lies between Supervised and
Unsupervised techniques. We use these
II. Classifications techniques when we are dealing with a data
which is a little bit labelled and rest large
Machine Learning is a new trending field portion of it is unlabelled. We can use
these days and is an application of artificial unsupervised technique to predict labels
intelligence. Machine learning uses certain and then feed these labels to supervised
statistical algorithms to make computers techniques. This technique is mostly
work in a certain way without being applicable in case of image data-sets where
explicitly programmed. The algorithms usually all images are not labelled.
receive an input value and predict an output REINFORCEMENT LEARNING, this
for this by the use of certain statistical type of learning uses three components
methods. The main aim of machine learning namely – agent, environment, action. An
is to create intelligent machines which can agent is the one that perceives its
think and work like human beings. On basis surroundings, an environment is the one
of the nature of the learning “signal” or with which an agent interacts and acts in
“feedback” available to a learning system, that environment. The main goal in
there are several types of machine learning reinforcement learning is to find the best
problems. possible policy.


indicates a presence of supervisor as III. Benefits
LABEL. Basically, supervised learning is a
learning in which we teach or train the 1.As machine learning has many wide
machine using data which is well labelled applications. Such as banking and financial
that means some data is already tagged with sector, healthcare, retail, publishing etc.
correct answer. After that, machine is 2. Google and Facebook are using machine
provided with new set of examples(data) so learning to push relevant advertisements.
That advertisements are based on users’ 3. We can say uses of machine algorithm is
past search behaviour. limited. Also, it’s not having any surety that
it’s algorithms will always work in every
3. Machine learning is used to handle
case imaginable. As we have seen that in
multi-dimensional and multi-variety data in
most cases machine learning fails. Thus, it
dynamic environments.
requires some understanding of the problem
4. Machine learning allows time cycle at hand to apply the right algorithm.
reduction and efficient utilization of
4. Like deep learning algorithm, machine
learning also needs a lot of training data. As
5. If one wants to provide continuous we can say it might be cumbersome to work
quality, large and complex process with a large amount of data. Fortunately,
environments. There are some tools present there are a lot of training data for image
because of machine learning. recognition purposes.

6. As there are too many things that come 5. There are fewer possibilities to make
under the practical benefit of machine immediate predictions with a machine
learning. Also, they involve the learning system. Also, don’t forget that it
development of autonomous computers, learns through historical data. Thus, the
software programs. Hence, it includes bigger the data and the longer it needs
processes that can lead to the automation of to expose to these data, the better it will
tasks. perform.

7. Fast Processing and Real-Time 9. No design guidelines. Accuracy depends

Predictions. on training and learning which is not always
8. The process of automation of tasks is
easily possible. 10. Hard to maintain degree of
9. Machine Learning helps user to utilize
the resource more efficiently by using 11. Hard to combine cases together.
previous processed data. Predictions are limited to the cases that
have been observed.
12. Required accurate details on many past
IV. Drawbacks projects.
1.Machine learning has the major challenge 13. Have large data requirement to learn
called Acquisition. Also, based on different about various topics which may be time
algorithms data need to be processed. And, taking and cause various resources.
it must be processed before providing as
input to respective algorithms. Thus, it has
a significant impact on results to be V. Applications
achieved or obtained.
As we move forward into the digital age,
2. As we have one more term interpretation.
one of the modern innovations we’ve seen
That it results is also a major challenge.
is the creation of Machine Learning. This
That need to determine the effectiveness of
incredible form of artificial intelligence is
machine learning algorithms.
already being used in various industries and
professions. For Example, Image and
Speech Recognition, Medical Diagnosis, We can segment the signal into portions
Prediction, Classification, Learning that contain distinct words or phonemes. In
Associations, Statistical each segment, we can represent the speech
Arbitrage, Extraction, Regression. Today signal by the intensities or energy in
we’re looking at all these Machine different time-frequency bands. Although
Learning Applications in today’s modern the details of signal representation are
world. These are the real-world Machine outside the scope of this program, we can
Learning Applications, represent the signal by a set of real values.
Speech recognition, Machine Learning
i. Image Recognition
applications include voice user interfaces.
It is one of the most common machine Voice user interfaces are such as voice
learning applications. There are many dialling, call routing, domestic appliance
situations where you can classify the object control. It can also use as simple data entry,
as a digital image. For digital images, the preparation of structured documents,
measurements describe the outputs of each speech-to-text processing, and plane.
pixel in the image. In the case of a black and
iii. Medical Diagnosis
white image, the intensity of each pixel
serves as one measurement. So, if a black ML provides methods, techniques, and
and white image has N*N pixels, the total tools that can help in solving diagnostic and
number of pixels and hence measurement is prognostic problems in a variety of medical
N2. In the coloured image, each pixel domains. It is being used for the analysis of
considered as providing 3 measurements of the importance of clinical parameters and of
the intensities of 3 main colour components their combinations for prognosis, e.g.
i.e. RGB. So, N*N coloured image there are prediction of disease progression, for the
3 N2 measurements. For face detection extraction of medical knowledge for
– The categories might be face versus no outcomes research, for therapy planning
face present. There might be a separate and support, and for overall patient
category for each person in a database of management. ML is also being used
several individuals. For character for data analysis, such as detection of
recognition – We can segment a piece of regularities in the data by appropriately
writing into smaller images, each dealing with imperfect data, interpretation
containing a single character. The of continuous data used in the Intensive
categories might consist of the 26 letters of Care Unit, and for intelligent alarming
the English alphabet, the 10 digits, and resulting in effective and efficient
some special characters. monitoring. It is argued that the successful
implementation of ML methods can help
ii. Speech Recognition
the integration of computer-based systems
Speech recognition (SR) is the translation in the healthcare environment providing
of spoken words into text. It is also known opportunities to facilitate and enhance the
as “automatic speech recognition” (ASR), work of medical experts and ultimately to
“computer speech recognition”, or “speech improve the efficiency and quality of
to text” (STT). In speech recognition, a medical care. In medical diagnosis, the
software application recognizes spoken main interest is in establishing the existence
words. The measurements in this Machine of a disease followed by its accurate
Learning application might be a set of identification. There is a separate category
numbers that represent the speech signal. for each disease under consideration and
one category for cases where no disease is screens and other devices. The image of the
present. Here, machine learning improves written text may be sensed "off line" from a
the accuracy of medical diagnosis by piece of paper by optical scanning (optical
analysing data of patients. The character recognition) or intelligent word
measurements in this Machine Learning recognition. Alternatively, the movements
applications are typically the results of of the pen tip may be sensed "on line", for
example by a pen-based computer screen
certain medical tests (example blood
surface, a generally easier task as there are
pressure, temperature and various blood more clues available. Handwriting
tests) or medical diagnostics, recognition principally entails optical
presence/absence/intensity of various character recognition. However, a complete
symptoms and basic physical information handwriting recognition system also
about the patient (age, sex, weight etc). On handles formatting, performs
the basis of the results of these correct segmentation into characters and
measurements, the doctors narrow down on finds the most plausible words.
the disease inflicting the patient. vi. Robot Locomotion
iv. Computer vision Robot locomotion is the collective name for
the various methods that robots use
Computer vision is an interdisciplinary to transport themselves from place to place.
scientific field that deals with how
computers can be made to gain high-level Wheeled robots are typically quite energy
understanding from digital efficient and simple to control. However,
other forms of locomotion may be more
images or videos. From the perspective
appropriate for a number of reasons, for
of engineering, it seeks to automate tasks
example traversing rough terrain, as well as
that the human visual system can do. moving and interacting in human
Computer vision tasks include methods environments. Furthermore, studying
for acquiring, processing, analysing and bipedal and insect-like robots may
understanding digital images, and beneficially impact on biomechanics. A
extraction of high-dimensional data from major goal in this field is in developing
the real world in order to produce numerical capabilities for robots to autonomously
or symbolic information, e.g., in the forms decide how, when, and where to move.
of decisions. Understanding in this context However, coordinating a large number of
means the transformation of visual images robot joints for even simple matters, like
(the input of the retina) into descriptions of negotiating stairs, is difficult. Autonomous
the world that can interface with other robot locomotion is a major technological
obstacle for many areas of robotics, such as
thought processes and elicit appropriate
action. This image understanding can be
seen as the disentangling of symbolic vii. Search engine
information from image data using models A web search engine or Internet search
constructed with the aid of geometry, engine is a software system that is designed
physics, statistics, and learning theory. to carry out web search, which means to
search the World Wide Web in a systematic
v. Handwriting recognition way for particular information specified in
Handwriting recognition (HWR) is the a web search query. The search results are
ability of a computer to receive and generally presented in a line of results,
interpret intelligible handwritten input often referred to as search engine results
from sources such pages . The information may be a mix
as paper documents, photographs, touch- of web pages, images, videos, infographics,
articles, research papers and other types of
files. Some search engines also mine
data available in databases or open
directories. Unlike web directories, which
are maintained only by human editors,
search engines also `maintain real-
time information by running
an algorithm on a web crawler. Internet
content that is not capable of being searched
by a web search engine is generally
described as the deep web.

VI. Conclusion
Here we discussed about the basics of
machine learning and its benefits along
with the drawbacks. We have also
discussed about the applications of
machine learning i.e. the role of machine
leaning in handwriting recognition,
medical diagnosis, speech and image
recognition, search engines and robot

VII. References
1. An Introduction to Machine
Learning - by Alexander J. Smola
2. https://www.quora.com/What-
3. https://www.profolus.com/topics/
4. https://www.google.com/amp/s/dat