See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/331428275

A Review of Artificial Intelligence Methods for Data Science and Data

Analytics: Applications and Research Challenges

Conference Paper · August 2018

DOI: 10.1109/I-SMAC.2018.8653670

CITATIONS READS

68 3,814

3 authors, including:

Krishna C V Mohana Mohana

BITS Pilani, Hyderabad R. V. College of Engineering
3 PUBLICATIONS   74 CITATIONS    74 PUBLICATIONS   1,396 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optoelectronic Properties of Graphene View project

All content following this page was uploaded by Mohana Mohana on 29 March 2021.

The user has requested enhancement of the downloaded file.

 Proceedings of the Second International conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2018)
IEEE Xplore Part Number:CFP18OZV-ART; ISBN:978-1-5386-1442-6

A Review of Artificial Intelligence Methods for

Data Science and Data Analytics: Applications and
Research Challenges
Krishna C V, Rohit H R, Mohana
Telecommunication Engineering
R. V. College of Engineering
Bengaluru, India

Abstract— Artificial intelligence is a field which requires programming language may be used for data processing and
multidisciplinary expertise where the final goal is to automate all analysis. Data science is evolving at an exponential space due
the human activities that presently require human intelligence. to the availability of tools, technologies and resources. It
The major problem is to develop a method which works exactly addresses many real word problems with optimistic solution
the way how a human brain works. The architecture of artificial
[2] [3] [4].
intelligence must emphasize on evaluation and redesign the
nature of design process. Data science is also trending now and A. Machine Learning
analytically deals to solve complex problems. Data is divided into
smaller parts and its trends, behaviors are understood. The main Machine learning technology has been existence from the
problem in data science is to handle large quantities of data. early 1950’s. In 1990 data driven approach is transformed in
Though there is significant increase in terms of research to machine learning. During the time period of 1995 to 2005
opportunities few challenges like lack of compute power, people there was a shift of focus towards the natural language search
power still remains a big challenge. and information retrieval. The neural network made a
comeback during 2005 where it was first tried in 1957 for first
Keywords—Deep Learning, Machine Learning, Artificial neural network computers. Machine Learning is among those
Intelligence, Data science and Data Analytics. technologies which have a number of successful and some
failure stories but in the near future (2 to 5 years) there are
I. INTRODUCTION possibilities of this technology to acquire mainstream. To keep
Artificial intelligence (AI) is the intelligence exhibited by up the growth in machine learning industry there should also
machines. Artificial intelligence is a method to simulate be growth in some of the factors affecting it such as
human intelligence using a set of algorithms and produce a infrastructure and technical skill
new machine which can do similar work with human B. Deep Learning
consciousness and also to perform parallel computing.
Machine learning is subset of artificial intelligence that gives In 1965 Deep Learning is introduced by Alexey Grigoryevich
the path towards designing of computers that are intelligent. Ivakhnenko and Valentin Grigor’ evich Lapa [4]. They used
Deep learning is a subset in machine learning, used to some models with polynomial functions and complicated
represent data abstraction though predefined model equations that was analyzed statically. In 1995, method for
architectures. Deep learning replicates the working of human recognizing and mapping related or similar data was
brain in data processing and creates patterns, reduces it if developed. During 1997 long short term memory for recurrent
possible and produces accurate results. This paper describes neural networks are established. During the late 1990’s with
the methods of AI, applications, hardware and software the development of the processors with fast computational
resources used and some of the research challenges. speeds, which eventually multiplied the computational speed
by an approximate factor of 1000 times which made the
processing of pictures by the GPU’s efficient. Different layers
II. EVOLUTION OF DATA SCIENCE AND DATA ANALYTICS of pre training and improvement of long short-term memory
The digital era of data science, was started in the mid-1900s, were used during the early 2000’s and by 2011, with the
where the Statistical Analysis System (SAS) was the growth of speed of GPUs,which made machines to work on
foundation of a project in North Carolina State University’s convolutional neural networks which doesn’t require the
(NCSU) agricultural department. When data science was first layer-by-layer pre-training. Currently the processing of Big
started in the industry level the main objective was to find Data is dependent on Deep Learning. Currently AI and Deep
more accurate and reliable solutions than those obtained using Learning growing and more advanced ideas are coming up.
business analysis. Data processing, predictive modeling, C. Artificial Intelligance (AI)
visualization, Sought Skillsets were those skills that an analyst In 1951 AI programs are written and run on Ferranti mark 1
must have to work on the data science field. In the present machine. AI research field initiated in 1956 at Dartmouth
time python and R are the main technologies used for data College during summer workshop [5]. During those times
processing [1]. However in the future Google’s Go there was the problem of computer hardware resources for

978-1-5386-1442-6/18/$31.00 ©2018 IEEE 591

 Proceedings of the Second International conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2018)
IEEE Xplore Part Number:CFP18OZV-ART; ISBN:978-1-5386-1442-6

computations. During the 1980’s billions of dollars were computes 30 trillion operations per second (TOPS) and
provided by the industry and government for the development consumes 30 watt of power. Intel AI Chip-Neural network
of AI after the insisting by the Government. During 2000 to processor provides flexible support for all deep learning
2010 after the funding and interest shown to develop the field primitives by making efficient core hardware components. It
of AI, there was a boom in the field. After the development of maximizes utilization of computation and scales multiple
powerful computer hardware machine learning became a compute nodes with less power.
successful approach to a number of problems in industry and
B. Software Resources
society.
Softwares and tools used for artificial intelligence and deep
III. ROLE OF ARTIFICIAL INTELLIGANCE IN DATA learning are explained.
ANALYTICS Pylearn2: designed for machine learning algorithms, flexible
Artificial intelligence and data analytics are hot research and extensible and it provides a library consisting of GPU and
topics in many parts of the world. Developing countries are CNN.
investing millions to gain prominence in world. Few of them Torch: It is AI and DL tool provides machine learning open
are investing a small amount in non-classified artificial source library. Scripting language used is Lua. Because of its
intelligence research. Artificial intelligence is a boon for flexibility and computational efficiency one of the most
industries in many sectors. popular tool. But the drawback is reimplementation logic is
not supported. Torchnet is the new open source framework, it
A. Machine Learning methods is reusable and efficient.
Decision trees uses variables or decision nodes in a hierarchy Theano: It is scripting language python library allows
and gives answers step by step. They are useful in evaluating evaluating mathematical expressions efficiently. It involves
lists of noticeable qualities, features. Support vector machine multi-dimensional arrays. Theano Faster than CPU in data-
is a supervised learning model that analyzes data used for intensive calculations because it uses mathematical
classification and regression analysis. They are good for expressions. For large data sets, using to train deep learning
binary classification of one variable versus other variables and networks [6].
there is no restriction for the relationship between variables to Caffe: It is an open source framework supports various
be linear. Naive Bayes classifier is based on concept libraries such as C++, python, MATLAB, CUDA for AI
probabilistic classifiers based on Bayes theorem with deployment architecture.
independence assumptions made between features. They Cuda-convnet: Fast neural network supports C++, CUDA and
compute the combined, conditional probabilities of multiple Python. It implements convolutional neural networks.
attributes. Deeplearning4j: It is an open source framework supports
various libraries such as C, C++, Java, Scala. It has a GPU
B. Deep Learning Methods support for distributed frame work library. Supports all most
Back propagation algorithm is used to calculate gradients all deep neural networks.
which are needed for weight calculation to be used by a TensorFlow: It is an open source framework for numerical
network. It has two modes of learning namely stochastic and computation. It is used where computations can be modeled as
batch. Dropout is a technique used for dropping the units data flow graph. It runs faster written in python. It has a
randomly during deep learning training. It is basically used for CUDA support, interfaces in C++, and also available on
reducing the number of parameters. Skip-gram is a method embedded platforms.
where if two vocabulary terms are similar only if they share Table I describes the comparision of various AI hardware and
similar context. For example given a word in middle of any software resources with respect to hardware utilization,
sentence it looks at words nearby and picks out one word at architecture, performance, platform and interfaces. Table II
random. The network will tell us the use of nearby word in our explains advantages and disadvantages of various AI methods.
vocabulary in terms of probability. Table III depicts the usage of AI in various applications [9]
[10] [11].
IV. SOFTWARE AND HARDWARE RESOURCES
V. TRENDS AND MARKET ANALYSIS
A. Hardware resources
Global AI market is expected to grow at a robust pace over the
For AI Technology many of researchers, data scientists and next five years, owing to its widespread implementation in
research organizations are using NVIDIA or INTEL hardware numerous industries, such as automobile, finance, healthcare,
GPU components for training, testing and to accelerate their consumer electronics, etc. Moreover, growing investments
deep learning algorithms. GPU performs parallel computing coupled with increase in number of acquisitions of start-ups by
tasks for data, images, videos and graphics efficiently with leading players such as Google, IBM, Microsoft and others
less duration. Nvidia Ai Chip-Nvidia’s new Jetson Xavier companies are playing a huge role in pushing the global
computer; it is compact size hardware contains various artificial intelligence market. During 2017 to 2022 the CAGR
processing components. It includes Volta tensor core GPU, 8 is expected to grow over 60% by the survey done by global AI
core ARM 64 CPU, 2 NVDLA accelerators and processors for market 2012-2022 [7]. As per the growth of global AI market
images and videos. It contains nine billion transistors and data analysis, NLP, audio and video processing, image

978-1-5386-1442-6/18/$31.00 ©2018 IEEE 592

 Proceedings of the Second International conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2018)
IEEE Xplore Part Number:CFP18OZV-ART; ISBN:978-1-5386-1442-6

recognition and gesture control has a major share. Among REFERENCESS

these categories, image recognition dominated global artificial [1] Data technology Landscape and evolution of data science .
https://www.digitalvidya.com.
intelligence market in 2016. And it is expected to continue the
[2] Evolution of data analytics: then, now and later-Affineblog.
dominance in next 5 years as well.
[3] The evolution of machine learning synectics for management decisions.
http://www.smdi.com.
VI. APPLICATIONS OF DATA SCIENCE AND DATA ANALYTICS
[4] A brief history of deep learning- Data versity.
[5] J Vincent , “ Nvidia launches AI computer to give autonomous robots
Today’s world has many applications of artificial intelligence better brains” the verge, 2018.
and research rate is increasing to make human life as simple as [6] Welcome – Theano 1.0.0 documentation.
possible. Most of the search engines use artificial intelligence http://deeplearning.net/software/theano/.
algorithms to predict the search by user. E-commerce sites [7] Global Artificial intelligance market size and trends.
uses artificial intelligence to recommend the users about http://www.techsciresearch.com.
products based on their recent searched products. Also the [8] V. Kumar and M. L. Garg, "Deep learning in predictive analytics: A
survey," 2017 International Conference on Emerging Trends in
delivery logistics companies use data science to improve their Computing and Communication Technologies (ICETCCT), Dehradun,
efficiency in operation. Websites like Junglee, Trivago and 2017, pp. 1-6.
many more are being driven by huge amount of data which is [9] P. Sherkhane and D. Vora, "Survey of deep learning software
fetched using RSS Feeds and APIs. Apart from the above few tools," 2017 International Conference on Data Management, Analytics
examples data science and artificial intelligence are used in and Innovation (ICDMAI), Pune, 2017, pp. 236-238.
every possible industry where data is generated and also where [10] J. Liu et al., "Applications of deep learning to MRI images: A survey,"
in Big Data Mining and Analytics, vol. 1, no. 1, pp. 1-18, March 2018.
data is analyzed.
[11] F. M. Hohman, M. Kahng, R. Pienta and D. H. Chau, "Visual Analytics
in Deep Learning: An Interrogative Survey for the Next Frontiers,"
VII. RESEARCH CHALLENGES OF DATA SCIENCE AND DATA in IEEE Transactions on Visualization and Computer Graphics.
ANALYTICS [12] Manjunath Jogin, Mohana, Madhulika, Meghana R K “Feature
extraction using Convolution Neural Networks (CNN) and Deep
A. Reasoning, problem solving Learning” 2018 IEEE International Conference On Recent Trends In
Electronics Information Communication Technology,(RTEICT) 2018,
During the late 1980’s and early 1990’s the algorithm used to India.
provide solutions for large reasoning problems were not [13] Arka Prava Jana, Abhiraj Biswas, Mohana, “YOLO based Detection and
sufficient because there were a lot of combinations possible Classification of Objects in video records” 2018 IEEE International
for a problem. This reduced the computational speeds to Conference On Recent Trends In Electronics Information
decrease exponentially as the problems grew larger and larger. Communication Technology,(RTEICT) 2018, India.
The concepts of probability and economics were the solutions [14] Chandan G, Ayush Jain, Harsh Jain, Mohana, “Real Time object
detection and Tracking using Deep Learning and openCV” International
developed by the researchers of AI to deal with uncertainty of Conference On Inventive research and computing applications
information or incomplete information. ,(ICIRCA) 2018, Coimbture,India.

B. Security
Security faults have been detected and studied in AI.An image
classifiers can be easily fooled by providing bad images or any
moving images or any kind of images which are perturbed
C. Motion and manipulation
AI robots when given a small static environment then it can
easily locate the place and map its environment but on the
other hand in a dynamic environment ,or where the movement
requires physical contact with the object then recognition of
those objects become difficult to program [5].
VIII. CONCLUSION
In current world we are advancing more towards artificial
intelligence and we are also witnessing the wide spread of it. It
is also important to realize the capacity of these technologies
in many ways that minimizes the risks safely. Machine
learning algorithms are now being used to predict analytics.
Deep learning is considered to have more advantages than the
conventional machine learning techniques in producing
accurate results. The more progress we make in this field, the
complexity and the challenges associated with it increases.

978-1-5386-1442-6/18/$31.00 ©2018 IEEE 593

 Proceedings of the Second International conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2018)
IEEE Xplore Part Number:CFP18OZV-ART; ISBN:978-1-5386-1442-6

TABLE I. COMPARISION OF ARTIFICIAL INTELLIGENCE HARDWARE AND SOFTWARE RESOURCES [9][12][13][14]

TABLE II COMPARISION OF ARTIFICIAL INTELLIGENCE HARDWARE AND SOFTWARE RESOURCES: ADVANTAGES AND DISADVANTAGES [9][12][13][14]

TABLE III COMPARISION OF ARTIFICIAL INTELLIGENCE HARDWARE AND SOFTWARE RESOURCES ADVANTAGES AND ITS USAGE IN APPLICATIONS
[9][12][13][14]

978-1-5386-1442-6/18/$31.00 ©2018 IEEE 594

View publication stats