Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

Semiconductor Science and Information Devices

https://ojs.bilpublishing.com/index.php/ssid

REVIEW
Machine Learning: A Review
Isonkobong Christopher Udousoro*
Department of Information Technology, School of Computing and Information Technology, Federal University of Tech-
nology, Owerri

ARTICLE INFO ABSTRACT

Article history Due to the complexity of data, interpretation of pattern or extraction of

Received: 28 May 2020 information becomes difficult; therefore application of machine learning
is used to teach machines how to handle data more efficiently. With the
Accepted: 2 July 2020 increase of datasets, various organizations now apply machine learning
Published Online: 31 October 2020 applications and algorithms. Many industries apply machine learning to
extract relevant information for analysis purposes. Many scholars, mathe-
Keywords: maticians and programmers have carried out research and applied several
Machine learning machine learning approaches in order to find solution to problems. In this
paper, we focus on general review of machine learning including various
Supervised learning machine learning techniques. These techniques can be applied to different
Unsupervised learning fields like image processing, data mining, predictive analysis and so on.
Reinforcement learning The paper aims at reviewing machine learning techniques and algorithms.
The research methodology is based on qualitative analysis where various
Semi-supervised learning
literatures is being reviewed based on machine learning.
Multitask learning
Ensemble learning
Neural Network
Instance-Based

1. Introduction ations can be handed over to machines [13]. Recently, with

Q
the introduction of artificial intelligence, machines have
uestions have been asked with regards to com- been created to have the same level of intelligence as the
puters if they are capable of learning on their human brains. Artificial intelligence started in 1943 when
own. Human beings have over the years created the first Neural Network Model was introduced [12].
different tools to enable them solve various tasks which A machine is expected to learn whenever there is chang-
led to the invention and production of different machines es in the structure, program or data, this is based on the in-
[57]
. With the rapid developments, the difference between put or response to the external environment which improves
humans and machines has remained intelligence. A human its expected results [51], therefore, machine learning can be
brain analyses information and makes decision according- defined as a part of artificial intelligence that explains that
ly but machines are not able to analyse and take decisions fact that machines can learn on their own when given the
[1]
. Automating tasks has generated high interest in the in- right data thereby solving a specific problem [80]. With the
formation technology field where some designs and oper- help of mathematics and statistics, machine learning can

*Corresponding Author:
Isonkobong Christopher Udousoro,
Department of Information Technology, School of Computing and Information Technology, Federal University of Technology, Owerri;
Email: [email protected]

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931 5

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

perform intellectual tasks independently that are always the go thereby improving the existing machine designs
generally performed by human beings [82]. because some designers produce non-workable machines
Machine learning is a part of computer science that that are not desired in the environment [50].
emanated from the study of pattern recognition and com- Knowledge computation is being carried out by ma-
putational learning theory all in artificial intelligence. Al- chines easily which will be a bit difficult for humans due
gorithms are used to make predictions on data [52]. Before to the large amount of knowledge available for certain
now the field of machine learning was mainly algorithms tasks [26]. Redesigning of systems due to change in en-
and theory of optimization but recently machine learning vironment is reduced with the introduction of machine
covers several other disciplines which includes statistics, learning because environments change overtime [51]. New
information theory, theory of algorithms, probability and knowledge, new task, new data is being gotten and dis-
functional analysis [24]. Machine learning and computa- covered by humans on every day, with machine learning,
tional statistics are always closely related because of their tracking of new knowledge is made easy [26].
specialty in prediction making and mathematical optimi- In this paper, a general overview of the application of
zation which brings about methods, theories and applica- machine learning is carried out. An introductory highlight
tion to the field [1]. In machine learning, strictly static pro- on the use of machine learning, its methods, techniques
gram instructions are not followed, rather, algorithms are and applications in various fields is considered. We also
used to build a model from input which are used to make carry out a survey on the existing work with machine
data-driven prediction or decisions [80]. learning so far. This paper is organized as follows: Section
Currently, research on machine learning focuses on ar- 2 describes machine learning, its applications and current
eas like pattern recognition, natural language processing, work being carried out. It also highlights the number of
cognitive computing, image processing, knowledge rep- journals and their sources being used for this paper. Sec-
resentation, traffic classification, cognitive radio and in- tion 3 introduces and highlights the techniques and meth-
trusion detection, optical networks and so on [13]. Machine ods used in machine learning while section 4 presents the
learning problems and tasks are mostly classified into conclusion of the paper.
three broad categories because of the signal and feedback
that is being fed into the learning system [53]. These cate- 2. Review of Literature
gories are explained below:
Machine learning is a fast growing field and one of the
(1) Supervised Learning where the computer is being
latest technologies being used recently in the information
supplied with example input and their desired output
technology field. Machine learning has been deployed to
which learns a rule that is able to map inputs to outputs [9]
solve different problems in different aspects of life like
(2) Unsupervised Learning where no label is given to
medical, engineering, agricultural etc. This section dis-
the learning algorithm and its allowed to determine the
cusses various applications of machine learning in differ-
structure of its input [70].
ent aspects of life.
(3) Semi-Supervised Learning lies between supervised
Machine learning is recently deployed in National
and unsupervised learning where the teacher gives an in-
Innovation Performance data analysis where the impact
complete input or signal where some of the target outputs
of government decisions and policies are still not clear.
are missing, it can also be regarded as Transduction [46].
A machine learning approach is proposed which includes
(4) Reinforcement Learning where an environment
clustering, correlation analysis, Bayesian Neural network
interacts with the computer to be able to perform a certain
and breakdown for decomposing innovation output pre-
goal without any input or interference of a teacher telling
diction. This approach has shown improvement for bench-
it what to do [28].
marking national innovation profiles [29].
There are various reasons why machine learning is im-
Machine learning is combined with satellite imaging
portant:
to predict poverty. A study carried out from five African
(1) Machine learning is important in adjusting its struc-
countries which are Nigeria, Tanzania, Uganda, Malawi
ture to produce desired outputs due to the heavy amount
and Rwanda. Here a neural network is used and trained
of data input into the system [57].
to capture and identify image features that explains local
(2) Machine learning is also suitable for data mining
economic outcomes [41].
because of the little amount of important data hidden in
Scientist are now called upon and required to use ad-
the heavy chunk of data that can be important for process-
vanced machine learning and many other artificial intelli-
ing of output [10].
gence technologies to help find new scientific discoveries
(3) Machine learning is important for jobs that are on
in the analysis of their data. This is important for applica-

6 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

tions like object recognition, natural language processing, cation domain for data mining and machine learning ap-
deep learning and automatic translation [39]. proaches. The advanced approach requires large training
Machine learning is applied in databases that are compiled set compounds to learn robust decision rules [50].
using symmetry-based calculations which contains thousands Wireless sensor networks work dynamically due to
of topological insulators and semimetals. Using the gradient the external factors or initiated by the system designers.
boosted trees approach of machine learning. The model pre- Therefore, machine learning is used to adapt to these con-
dicts the topology of a given existent material. Although the ditions and eliminate the need for unnecessary and also
model has errors which can be overcome [21]. prolong the lifespan of the network [1].
In the field of chemistry, a machine learning approach Quantum systems are said to be faster and produce pat-
known as graph neural networks is proposed for solving terns that classical systems cannot which means quantum
problem of quantitative structure-odor relationship which computers may outperform classical computers in terms
relationship between molecules structure and its odor re- of machine learning. Although hardware and software
mains difficult [78]. challenges still exist, quantum algorithms could act as a
Based on widely analysed classification model, patient building block for machine learning programs [6].
data is being analysed for the predictability of the patient In project management, project delay is one of the press-
to have liver disease. There are five phases highlighted. ing challenges faced by the construction sector due to com-
The first is the min-max algorithm application followed plexity and delay risk sources. Machine learning algorithms
by the use of PSO feature selection for demarcation of at- identified and developed in order to facilitate accurate delay
tributes, then classification algorithm is used for compar- risk analysis and prediction using objective data sources.
ative analysis and categorization. The fourth phase is the Two machine learning algorithms which are decision tree
accuracy calculation and finally evaluation phase [26]. and naive Bayesian classification were trained using the
Decision Tree based algorithm is used to evaluate in- data set for predicting project delay events [32].
dividual and ensemble model performance for predicting In order to minimize investment risks while evaluating
secondary progressive multiple sclerosis disability pro- the potential business impact, machine learning systems is
gression +. used to leverage business transformation [64].
Machine learning is a branch of artificial intelligence Retrieval of patients data rightly and at the right time
that is deployed and most suited for medical applications is vital, therefore, a learning electronic medical record
especially in the detection and prediction of cancer and (LEMR) is developed that learns statistical models of cli-
prognosis. A number of publications have shown that ma- nician information seeking behaviour and applies it to the
chine learning methods can be used to improve the accu- direct display of data for future patients [45].
racy of predicting cancer susceptibility, reoccurrence and
mortality [23]. 3. Machine Learning Techniques
The use of Bayesian deep learning method as machine
Machine learning is simply about designing algorithm
learning technique to address the analysis challenges of
that will allow a computer to be able to learn [52]. This is
future surveys. This technique will enable multibendpass,
based on the input and desired outcome of the algorithm.
multi-instrument processing of individual images, target-
Some of the machine learning techniques will define how
ing science objectives [4].
humans can approach a task [64]. Several mathematicians
In order to manage and control the operational environ-
and programmers have come up with solutions on the
ment while also predicting future actions, IoT application
approaches and techniques for machine learning which is
developers tend to buy data from IoT owners in order to
narrowed down as shown in the diagram below.
train machine learning models by combining edge com-
puting resources with data sources [76].
Machine learning is used in the multimedia world for Machine learning

conversion assignment. The audio/video data is recorded Supervised

learning
Unsupervised
learning
Semi-
supervised
learning
Reinforcement
learning
Multi-task
learning
Ensemble
learning
Neural Network
Instance
based
learning

from a location and converted to text data. The text data Decision Principal Generative
Boosting
Supervised
Neural
K-nearest
Models
is then analysed and tagged to each individual based on
Tree Component Network
neighbor
Analysis
Unsupervised
Bagging
attributes. Self learning software designed using machine
Naive Bayes Self Training Neural
K-Means Network

learning algorithm is generated based on the tagging [24].

Support Vector Transductive
Machine Support Vector Reinforcement
Machine Neural Network

Machine learning approach has also be deployed in the

discovery of drugs. The use of virtual screening evolving
Figure 1. Machine Learning Techniques [25]
from traditional similarity searching to an advanced appli-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931 7

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

The various techniques of machine learning is dis- (b) Naive Bayes: This algorithm is mostly used and
cussed below and the different applications that follow the is a target of the text classification industry [83]. It is also
techniques are also highlighted: used for clustering and classification purposes. Condition-
(1) SUPERVISED LEARNING: In supervised learning, al probability is the backbone of Naive bayes algorithm
a correct classification is already assigned to train a data where it creates trees based on the probability of occur-
sample from the data source [22]. It can also be seen as a for- ring. These trees can also be regarded as Bayesian Net-
malization of a certain idea of learning from examples where work. An example is shown below.
there is an input and desired output [25]. Here, the learner
which could be a computer program is provided with a train-
ing set and test set of data. The trainer is required to learn
from the training set with examples of labelled set which will
be used to identify the unlabelled examples in the test set
with highest possible accuracy [15]. The supervised learning
technique requires external assistance. The work flow of the
supervised learning technique is shown below.

Problem

Identification Figure 4. Bayesian Network [25]

of Required
data (c) Support Vector Machine: This algorithm is a com-
Data Pre-Processing monly and widely used machine learning technique and
mostly used for classification [90]. Support Vector Machine
Definition of Training set uses margin calculations where it draws margins between
Algorithm
the classes. The distance between margins and classes is
Selection always big so as to reduce the error in classification [3].
The diagram below shows a working support vector ma-
Parameter Training
chine.
Evaluation of Test set

OK? Classifier

Figure 2. Workflow of supervised machine learning tech-

nique [25]
There are three major types or algorithm of supervised
learning technique as discussed below:
(a) Decision Tree: Decision trees group attributes there-
by sorting them based on their different values [62]. Each of
this decision trees consist of nodes and branches and are
mainly used for classification. A node represents attributes
in a group while the branch represents a value that the node
can take. An example of a decision tree is shown below.

Age

<=30 >30
Figure 5. A working Support Vector Machine [25]
Gender
No An application of the supervised learning technique is
where a supervised data of the Stanford Natural Language
Male Female
Inference datasets is used to train and show how universal
sentence representations can consistently outperform un-
Yes No supervised methods [22].
(2) UNSUPERVISED LEARNING: This technique is a
Figure 3. Decision Tree [25]

8 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

bit harder than the supervised learning. This is so because vised learning technique are discussed below.
we tell the computer to learn to do something that we (a) Generative Models: This is one of the oldest
don’t tell it how to do [82]. This learning technique does not semi-supervised learning technique where models like
produce classification but make decisions that maximize guassian mixture models is used as a mixed distribution to
rewards. Some self-organized neural networks learn using assume a structure [48]. The mixed component can be iden-
the unsupervised learning technique to uncover hidden tified within the unlabelled data.
patterns in unlabelled data input [89]. The advantage of this (b) Self-Training: Here, the classified is trained to
lack of direction is that it lets the algorithm to look back learn by itself by providing it with portion of labelled
for patterns that were not previously considered. The un- data where the labelled points and the predicted labels are
supervised learning technique learns few features from a added together in the training set, and then the process is
set of data and then when a new data is being introduced, repeated [65].
then it uses the learned features to recognize the class of (c) Transductive Support Vector Machine: This algo-
the data [70]. Unsupervised learning technique is mainly rithm is an extension of the Support Vector Machine algo-
used for clustering and feature reduction. And example is rithm where both the labelled and unlabelled data is con-
shown below. sidered [86]. This algorithm is used to label the unlabelled
data such that the margin is large between the labelled and
unlabelled data.
Odena Augustus in 2015 extended Generative adver-
sarial networks to the semi-supervised learning technique
of machine learning to show it can be used to create a
more data-efficient classifier and higher quality samples
can be generated. Semi-supervised learning is applied on
graph-structured data which learns hidden layer represen-
tation that encode the graph structure and feature of nodes
Figure 6. Example of Unsupervised Learning [25] [46]
.
There are two main algorithms for supervised learning (4) REINFORCEMENT LEARNING: In this type of
technique which are discussed below: technique, the positive outcome of the decisions is deter-
(a) K-Means Clustering: Clustering is also known as minant or dependent on the actions to take [95]. The learner
grouping which is a type of unsupervised learning tech- has no idea of the action to take until it is given a particu-
nique that creates group which initiated automatically [87]. lar situation. Depending on the actions taken by the learn-
It is known as K-means because it creates k-distance clus- er, the future is affected in terms of the situations. Below
ters where items with similar characteristics are put in the shows a model for reinforcement learning.
same cluster. The centre of the cluster is gotten from the
mean of the values in the cluster.
(b) Principal Component Analysis: Here, to make the
computation faster and a bit easier, the dimension of the
data is reduced. For example, when principal component
analysis is being applied on a 2D data, the data will then
be reduced to 1D [87].
Unsupervised learning techniques can be applied to
sentences to improve embeddings [72]. Unsupervised learn-
ing technique can be used to identify phases and its tran-
sitions in systems, principal component analysis is used to
extract the original data while clustering is used to identi-
fy the phases [87].
(3) SEMI SUPERVISED LEARNING: This technique Figure 7. Reinforcement Learning Model [25]
is a technique that combines both supervised and unsu-
In the above model, the input i, is being received by
pervised learning techniques [46]. The unlabelled data is
the agent. The agent also receives current state, s, state
already present while discovering the labelled data is very
transition r, and input function I from the environment.
tasking and tedious. This technique is common in data
With these inputs, the agent generates a behaviour B and
mining field [71]. Some of the algorithms for semi-super-
takes an action a which generates an outcome [25]. Rein-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931 9

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

forcement learning technique is being applied to natural The input layer takes input while the hidden layer pro-
language processing for dialogue creation where a model cesses the input and then the output layer sends the calcu-
simulates dialogues between vitual agents using policy lated output. Artificial Neural Network can be divided into
gradients for reward to conversational properties [54]. three types which are supervised, unsupervised and rein-
(5) MULTITASK LEARNING: The algorithm remem- forcement neural network. Neural networks are powerful
bers the process and procedure how a particular problem and works well for difficult learning task like in image,
was solved and how it arrived at a certain conclusion speech and natural language processing [95].
[54]
. The process and procedure is being used to proffer (8) INSTANCE-BASED LEARNING: In this tech-
solution to other tasks or problem. It can also be known nique, the learner is familiar with only one particular
as transfer mechanism [73]. Sharing experience between type of pattern where it tries to apply to newly fed data
[16]
learners helps them to learn concurrently rather than in- . The technique is termed lazy because it waits for the
dividually which is much faster. Four clinical predictions test data to arrive and then act on it with training data. It
benchmark is proposed using data derived from publicly gets complex as the data increases. An example of the in-
available medical information database where multitask stance-based learning technique is the k-nearest neighbour
learning, deep supervision and data specific modifications algorithm. In the agricultural field, the instance based
is applied on the performance of neural models [36]. learning approach is used to segment crop images where-
(6) ENSEMBLE LEARNING: This technique refers to by green texture crops are automatically discriminated
the combination of individual learners to form one learner from the rest of the crops [2].
[49]
. For example, a decision tree, naive baye and neural
network can be combined to form an ensemble learning. 4. Conclusion
Combination of learners performs better work than indi-
This review presented a general research on machine
vidual learner. There are two algorithms under ensemble
learning, its algorithm and techniques. The paper reviews
learning:
literatures on the applications of machine learning al-
(a) Boosting: This technique collects weak learners and
gorithm and techniques on various fields of life which
converts them to one strong learner [92]. It decreases bias
include medical, agriculture, science and so on. Machine
and variance.
learning is one of the high rising technologies used re-
(b) Bagging: This technique is also known as boostrap
cently for solutions to various problems. It has various
aggregating, whenever the machine learning algorithm
algorithms which include supervised, semi-supervised,
needs to increase accuracy and stability, bagging tech-
unsupervised, reinforcement algorithms and so on.
nique is required [92].
Machine learning models are built and proposed us- References
ing ensemble learning technique where large repository
of malware samples and applications are gotten from an [1] Alsheikh, M. A., Lin, S., Niyato, D., Tan, H. P. Ma-
anti-virus vendor. This is in vie to reduce that malware chine learning in wireless sensor networks: Algo-
attack on the android platform [92] rithms, strategies, and applications. IEEE Communi-
(7) NEURAL NETWORK LEARNING: This tech- cations Surveys & Tutorials, 2014, 16(4): 1996-2018.
nique is gotten from the concept of neurons which has [2] Arroyo, J., Guijarro, M., Pajares, G. An in-
four parts, dendrites, nucleus, Soma and Axon[58]. Inter- stance-based learning approach for thresholding in
connection of the neurons is known as neural network. crop images under different outdoor conditions. Com-
puters and Electronics in Agriculture, 2016, 127:
The artificial neural network behaves in the same pattern.
669-679.
A diagram of artificial neural network is shown below.
[3] Ashfaq, R. A. R., Wang, X. Z., Huang, J. Z., Abbas,
H., He, Y. L. Fuzziness based semi-supervised learn-
ing approach for intrusion detection system. Informa-
tion Sciences, 2017, 378: 484-497.
[4] Aubourg, É., Bartlett, J., Boucaud, A., Ganga, K.,
Giraud-Héraud, Y., Le Jeune, M., LAL, J. É. C.
Prospective IN2P3 Survey Synergies with Machine
Learning GT05+ GT09, 2019.
[5] Bakoev, S., Getmantseva, L., Kolosova, M., Kostyu-
nina, O., Chartier, D., Tatarinova, T. V. PigLeg: Pre-
Figure 8. Artificial Neural Network [25] diction of Swine Phenotype Using Machine Learn-

10 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

ing, 2019. Automatic Detection of Microlensing Events in the

[6] Balzer, L. B., Havlir, D. V., Kamya, M. R., Chamie, Galactic Bulge using Machine Learning Techniques.
G., Charlebois, E. D., Clark, T. D., Kabami, J. Ma- In Astronomical Data Analysis Software and Systems
chine learning to identify persons at high-risk of HIV XXVII, 2019, 523: 127.
acquisition in rural Kenya and Uganda. Clinical In- [21] Claussen, N., Bernevig, B. A., Regnault, N. Detec-
fectious Diseases, 2019. tion of Topological Materials with Machine Learn-
[7] Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., ing. arXiv preprint arXiv:1910.10161, 2019.
Wiebe, N., Lloyd, S. Quantum machine learning. Na- [22] Conneau, A., Kiela, D., Schwenk, H., Barrault, L.,
ture, 2017, 549(7671): 195-202. Bordes, A. Supervised learning of universal sentence
[8] Bojanowski, P., Joulin, A. Unsupervised learning by representations from natural language inference
predicting noise. In Proceedings of the 34th Interna- data. arXiv preprint arXiv:1705.02364, 2017.
tional Conference on Machine Learning. JMLR. Org, [23] Cruz, J. A., Wishart, D. S. Applications of machine
2017, 70: 517-526. learning in cancer prediction and prognosis. Cancer
[9] Bostanabad, R., Bui, A. T., Xie, W., Apley, D. W., informatics, 2, 117693510600200030, 2006.
Chen, W. Stochastic microstructure characterization [24] Dawson, C. J., Molloy, C. L., Trim, C. M., Ganci Jr, J.
and reconstruction via supervised learning. Acta Ma- M. U.S. Patent Application No. 15/952,320, 2019.
terialia, 2016, 103: 89-102. [25] Dey, A. Machine learning algorithms: a review. Inter-
[10] Brownlee, J. Machine learning mastery, 2014. URL: national Journal of Computer Science and Informa-
http://machinelearningmastery.com/discover-feature-en- tion Technologies, 2016, 7(3): 1174-1179.
gineering-howtoengineer-features-and-how-to-get- [26] Durai, V., Ramesh, S., Kalthireddy, D. Liver disease
good-at-it prediction using machine learning, 2019.
[11] Butler, K. T., Davies, D. W., Cartwright, H., Isayev, [27] Durand, T., Mordan, T., Thome, N., Cord, M. Wild-
O., Walsh, A. Machine learning for molecular and cat: Weakly supervised learning of deep convnets for
materials science. Nature, 2018, 559(7715): 547. image classification, pointwise localization and seg-
[12] Chaker, Z., Salanne, M., Delaye, J. M., Charpentier, T. mentation. In Proceedings of the IEEE conference on
Fast and accurate predictions of NMR parameters in computer vision and pattern recognition, 2017: 642-
aluminosilicate glasses via Machine Learning, 2019. 651.
[13] Chatterjee, D., Ghosh, S., Brady, P. R., Kapadia, S. J., [28] Foerster, J., Assael, I. A., de Freitas, N., Whiteson,
Miller, A. L., Nissanke, S., Pannarale, F., 2019. S. Learning to communicate with deep multi-agent
[14] Chen, D., Mak, B. K. W. Multitask learning of deep reinforcement learning. In Advances in Neural Infor-
neural networks for low-resource speech recogni- mation Processing Systems, 2016: 2137-2145.
tion. IEEE/ACM Transactions on Audio, Speech, and [29] Forner, D., Ozcan, S., Bacon, D. Machine Learning
Language Processing, 2015, 23(7): 1172-1183. Approach for National Innovation Performance Data
[15] Chen, X., Gupta, A. Webly supervised learning of Analysis, 2019.
convolutional networks. In Proceedings of the IEEE [30] Fu, G. S., Levin-Schwartz, Y., Lin, Q. H., Zhang, D.
International Conference on Computer Vision, 2015: Machine Learning for Medical Imaging. Journal of
1431-1439. healthcare engineering, 2019.
[16] Cheng, M. Y., Hoang, N. D. A Swarm-Optimized [31] Gales, M. Introduction to machine learning, 2008.
Fuzzy Instance-based Learning approach for pre- [32] Gondia, A., Siam, A., El-Dakhakhni, W., Nassar, A.
dicting slope collapses in mountain roads. Knowl- H. Machine Learning Algorithms for Construction
edge-Based Systems, 2015, 76: 256-263. Projects Delay Risk Prediction. Journal of Construc-
[17] Cheng, Y. Semi-supervised learning for neural ma- tion Engineering and Management, 2019, 146(1):
chine translation. In Joint Training for Neural Ma- 04019085.
chine Translation. Springer, Singapore, 2019: 25-40. [33] Greenberg, D. E., Kim, J., Zhan, X., Shelburne, S. A.,
[18] Chikersal, P., Poria, S., Cambria, E. SeNTU: senti- Shelburne, S. A., Aitken, S. L., Aitken, S. L. 1831.
ment analysis of tweets by combining a rule-based Machine Learning Approaches to Predicting Resis-
classifier with supervised learning. In Proceedings of tance in Pseudomonas aeruginosa. In Open Forum
the 9th International Workshop on Semantic Evalua- Infectious Diseases. Oxford University Press, 2019,
tion (SemEval 2015), 2015: 647-651. 6(Suppl 2): S42.
[19] Chong, B. S. A machine learning approach to detect [34] Gu, S., Holly, E., Lillicrap, T., Levine, S. Deep re-
surface features for automatic robot taping, 2019. inforcement learning for robotic manipulation with
[20] Chu, S., Wagstaff, K., Bryden, G., Shvartzvald, Y. asynchronous off-policy updates. In 2017 IEEE in-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931 11

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

ternational conference on robotics and automation Optimization. arXiv preprint arXiv:1910.12741,

(ICRA). IEEE, 2017: 3389-3396. 2019.
[35] Hamilton, I. Fraud Detection Through the Utilization [48] Kuznietsov, Y., Stuckler, J., Leibe, B. Semi-super-
of Machine Learning. 2019. vised deep learning for monocular depth map pre-
[36] Harutyunyan, H., Khachatrian, H., Kale, D. C., diction. In Proceedings of the IEEE Conference on
Steeg, G. V., Galstyan, A. Multitask learning and Computer Vision and Pattern Recognition, 2017:
benchmarking with clinical time series data. arXiv 6647-6655.
preprint arXiv:1703.07771, 2017. [49] Laradji, I. H., Alshayeb, M., Ghouti, L. Software
[37] Harwath, D., Torralba, A., Glass, J. Unsupervised defect prediction using ensemble learning on select-
learning of spoken language with visual context. ed features. Information and Software Technolo-
In Advances in Neural Information Processing Sys- gy, 2015, 58: 388-402.
tems, 2016: 1858-1866. [50] Lavecchia, A. Machine-learning approaches in drug
[38] He, B., Wei, M., Watts, D. R., Donohue, K. A., Trac- discovery: methods and applications. Drug discovery
ey, K. L., Shen, Y. Detect slow slip events in ocean today, 2015, 20(3): 318-331.
bottom pressure data using machine learning. Thresh- [51] Law, M. T., Traboulsee, A. L., Li, D. K., Carruthers,
old, 2019, 68: 95-5. R. L., Freedman, M. S., Kolind, S. H., Tam, R. Ma-
[39] Hey, T., Butler, K., Jackson, S., Thiyagalingam, J. chine learning in secondary progressive multiple
Machine Learning and Big Scientific Data. arXiv sclerosis: an improved predictive model for short-
preprint arXiv:1910.07631, 2019. term disability progression. Multiple Sclerosis Jour-
[40] Hong, S., You, T., Kwak, S., Han, B. Online tracking nal–Experimental, Translational and Clinical, 2019,
by learning discriminative saliency map with convo- 5(4): 2055217319885983.
lutional neural network. In International conference [52] Learned-Miller, E. G. Introduction to supervised
on machine learning, 2015: 597-606. learning. I: Department of Computer Science, Uni-
[41] Jean, N., Burke, M., Xie, M., Davis, W. M., Lobell, versity of Massachusetts, 2014.
D. B., Ermon, S. Combining satellite imagery and [53] Lemm, S., Blankertz, B., Dickhaus, T., Müller, K.
machine learning to predict poverty. Science, 2016, R. Introduction to machine learning for brain imag-
353(6301): 790-794. ing. Neuroimage, 2011, 56(2): 387-399.
[42] Jean, N., Burke, M., Xie, M., Davis, W. M., Lobell, [54] Li, J., Monroe, W., Ritter, A., Galley, M., Gao, J., Ju-
D. B., Ermon, S. Combining satellite imagery and rafsky, D. Deep reinforcement learning for dialogue
machine learning to predict poverty. Science, 2016, generation. arXiv preprint arXiv:1606.01541, 2016.
353(6301): 790-794. [55] Liang, X., Liu, S., Wei, Y., Liu, L., Lin, L., Yan, S.
[43] Johnson, R., Zhang, T. Semi-supervised convolution- Towards computational baby learning: A weakly-su-
al neural networks for text categorization via region pervised approach for object detection. In Proceed-
embedding. In Advances in neural information pro- ings of the IEEE International Conference on Com-
cessing systems, 2015: 919-927. puter Vision, 2015: 999-1007.
[44] Khasawneh, K. N., Ozsoy, M., Donovick, C., [56] Lin, E., Tsai, S. J. Machine Learning in Neural Net-
Abu-Ghazaleh, N., & Ponomarev, D. Ensemble works. In Frontiers in Psychiatry. Springer, Singa-
learning for low-level hardware-supported malware pore, 2019: 127-137.
detection. In International Symposium on Recent Ad- [57] Lison, P. An introduction to machine learning. Lan-
vances in Intrusion Detection. Springer, Cham, 2015: guage Technology Group (LTG), 2015, 1(35).
3-25. [58] Liu, T., Yang, Y., Huang, G. B., Yeo, Y. K., Lin, Z.
[45] King, A. J., Cooper, G. F., Clermont, G., Hochheiser, Driver distraction detection using semi-supervised
H., Hauskrecht, M., Sittig, D. F., Visweswaran, S. machine learning. IEEE transactions on intelligent
Using Machine Learning to Selectively Highlight Pa- transportation systems, 2015, 17(4): 1108-1120.
tient Information. Journal of Biomedical Informatics, [59] Liu, Y. J., Li, J., Tong, S., Chen, C. P. Neural network
2019: 103327. control-based adaptive learning design for nonlinear
[46] Kipf, T. N., Welling, M. Semi-supervised classifica- systems with full-state constraints. IEEE transac-
tion with graph convolutional networks. arXiv pre- tions on neural networks and learning systems, 2016,
print arXiv:1609.02907, 2016. 27(7): 1562-1571.
[47] Kudyshev, Z. A., Kildishev, A. V., Shalaev, V. M., [60] Liu, Y. Y., Welch, D., England, R., Stacey, J., Har-
Boltasseva, A. Machine-Learning-Assisted Meta- bison, S. Forensic STR allele extraction using a
surface Design for High-Efficiency Thermal Emitter machine learning paradigm. Forensic Science Inter-

12 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

national: Genetics, 102194, 2019. 2017.

[61] Mairal, J. Incremental majorization-minimization [73] Peng, H., Thomson, S., Smith, N. A. Deep multitask
optimization with application to large-scale machine learning for semantic dependency parsing. arXiv pre-
learning. SIAM Journal on Optimization, 2015, print arXiv:1704.06855, 2017.
25(2): 829-855. [74] Pyati, V. U.S. Patent Application No. 15/949,107,
[62] Makantasis, K., Karantzalos, K., Doulamis, A., Dou- 2019.
lamis, N. Deep supervised learning for hyperspectral [75] Rezende, D. J., Eslami, S. A., Mohamed, S., Batt-
data classification through convolutional neural net- aglia, P., Jaderberg, M., Heess, N. Unsupervised
works. In 2015 IEEE International Geoscience and learning of 3d structure from images. In Advances in
Remote Sensing Symposium (IGARSS). IEEE, 2015: Neural Information Processing Systems, 2016: 4996-
4959-4962. 5004.
[63] Masood, H., Toe, C. Y., Teoh, W. Y., Sethu, V., Amal, [76] Sajan, K. K., Ramachandran, G. S., Krishnamach-
R. Machine Learning for Accelerated Discovery of ari, B. Enhancing Support for Machine Learning
Solar Photocatalysts. ACS Catalysis, 2019. and Edge Computing on an IoT Data Marketplace.
[64] Mattos, D. I., Bosch, J., Olsson, H. H. Leveraging In Proceedings of the First International Workshop
Business Transformation with Machine Learning Ex- on Challenges in Artificial Intelligence and Machine
periments. In International Conference on Software Learning for Internet of Things. ACM, 2019: 19-24.
Business. Springer, Cham, 2019: 183-191. [77] Salamon, J., Bello, J. P. Unsupervised feature learn-
[65] Misra, I., Shrivastava, A., Hebert, M. Watch and ing for urban sound classification. In 2015 IEEE
learn: Semi-supervised learning for object detectors International Conference on Acoustics, Speech and
from video. In Proceedings of the IEEE Conference Signal Processing (ICASSP). IEEE, 2015: 171-175.
on Computer Vision and Pattern Recognition, 2015: [78] Sanchez-Lengeling, B., Wei, J. N., Lee, B. K., Ger-
3593-3602. kin, R. C., Aspuru-Guzik, A., Wiltschko, A. B. Ma-
[66] Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. chine Learning for Scent: Learning Generalizable
A., Veness, J., Bellemare, M. G., Petersen, S. Hu- Perceptual Representations of Small Molecules. arX-
man-level control through deep reinforcement learn- iv preprint arXiv:1910.10685, 2019.
ing. Nature, 2015, 518(7540): 529. [79] Sanchez-Lengeling, B., Wei, J. N., Lee, B. K., Ger-
[67] Munro, R. J., Walker, C., Luger, S. K., Callahan, B. kin, R. C., Aspuru-Guzik, A., & Wiltschko, A. B.
D., King, G. C., Tepper, P. A., Long, J. D. U.S. Patent Machine Learning for Scent: Learning Generalizable
Application No. 16/185,843, 2019. Perceptual Representations of Small Molecules. arX-
[68] Murthy, A., Green, C., Stoleru, R., Bhunia, S., Swan- iv preprint arXiv:1910.10685, 2019.
son, C., Chaspari, T. Machine Learning-based Irriga- [80] Schuld, M., Sinayskiy, I., Petruccione, F. An intro-
tion Control Optimization. In Proceedings of the 6th duction to quantum machine learning. Contemporary
ACM International Conference on Systems for En- Physics, 2015, 56(2): 172-185.
ergy-Efficient Buildings, Cities, and Transportation. [81] Shen, W. U.S. Patent No. 10,467,339. Washington,
ACM, 2019: 213-222. DC: U.S. Patent and Trademark Office, 2019.
[69] Narayanaswamy, S., Paige, T. B., Van de Meent, J. [82] Srivastava, N., Mansimov, E., Salakhudinov, R. Un-
W., Desmaison, A., Goodman, N., Kohli, P., Torr, P. supervised learning of video representations using
Learning disentangled representations with semi-su- lstms. In International conference on machine learn-
pervised deep generative models. In Advances in ing, 2015: 843-852.
Neural Information Processing Systems, 2017: 5925- [83] Stoudenmire, E., Schwab, D. J. Supervised learning
5935. with tensor networks. In Advances in Neural Infor-
[70] Noroozi, M., Favaro, P. Unsupervised learning of vi- mation Processing Systems, 2016: 4799-4807.
sual representations by solving jigsaw puzzles. In Eu- [84] Sun, Y., Tang, K., Minku, L. L., Wang, S., Yao, X.
ropean Conference on Computer Vision. Springer, Online ensemble learning of data streams with grad-
Cham, 2016: 69-84 . ually evolved classes. IEEE Transactions on Knowl-
[71] Odena, A. Semi-supervised learning with gener- edge and Data Engineering, 2016, 28(6): 1532-1545.
ative adversarial networks. arXiv preprint arX- [85] Tandia, A., Onbasli, M. C., Mauro, J. C. Machine
iv:1606.01583, 2016. Learning for Glass Modeling. In Springer Handbook
[72] Pagliardini, M., Gupta, P., Jaggi, M. Unsupervised of Glass. Springer, Cham, 2019: 1157-1192.
learning of sentence embeddings using composition- [86] Tu, Y., Lin, Y., Wang, J., Kim, J. U. Semi-supervised
al n-gram features. arXiv preprint arXiv:1703.02507, learning with generative adversarial networks on dig-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931 13

 Semiconductor Science and Information Devices | Volume 02 | Issue 02 | October 2020

ital signal modulation classification. Comput. Mater. ing. IET Information Security, 2015, 9(6): 313-320.
Continua, 2018, 55(2): 243-254. [92] Zhang, F., Du, B., Zhang, L., Xu, M. Weakly super-
[87] Wang, L.Discovering phase transitions with unsu- vised learning based on coupled convolutional neural
pervised learning. Physical Review B, 2016, 94(19): networks for aircraft detection. IEEE Transactions on
195105. Geoscience and Remote Sensing, 2016, 54(9): 5553-
[88] Wang, S., Yin, Y., Cao, G., Wei, B., Zheng, Y., Yang, 5563
G. Hierarchical retinal blood vessel segmentation [93] Zhao, M., Chow, T. W., Zhang, Z., Li, B. Automatic
based on feature and ensemble learning. Neurocom- image annotation via compact graph based semi-su-
puting, 2015, 149: 708-717. pervised learning. Knowledge-Based Systems, 2015,
[89] Wang, X., Gupta, A. Unsupervised learning of visual 76: 148-165.
representations using videos. In Proceedings of the [94] Zhao, Y., Li, J., Yu, L. A deep learning ensemble
IEEE International Conference on Computer Vision, approach for crude oil price forecasting. Energy Eco-
2015: 2794-2802. nomics, 2017, 66: 9-16.
[90] Yao, X., Han, J., Cheng, G., Qian, X., Guo, L. Yang, [95] Zoph, B., Le, Q. V. Neural architecture search
H. F., Lin, K., Chen, C. S. Supervised learning of se- with reinforcement learning. arXiv preprint arX-
mantics-preserving hash via deep convolutional neu- iv:1611.01578,2016.
ral networks. IEEE transactions on pattern analysis [96] Ren, Z., Yan, J., Ni, B., Liu, B., Yang, X., Zha, H.
and machine intelligence, 2017, 40(2): 437-451. Unsupervised deep learning for optical flow estima-
[91] Yerima, S. Y., Sezer, S., Muttik, I. High accuracy tion. In Thirty-First AAAI Conference on Artificial
android malware detection using ensemble learn- Intelligence, 2017.

14 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/ssid.v2i2.1931