The upsurge of deep learning for computer vision applications

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 10, No. 1, February 2020, pp. 538~548
ISSN: 2088-8708, DOI: 10.11591/ijece.v10i1.pp538-548  538
Journal homepage: http://ijece.iaescore.com/index.php/IJECE
The upsurge of deep learning for computer vision applications
Priyanka Patel1
, Amit Thakkar2
1
U & P U Patel Department of Computer Engineering, Chandubhai S Patel Institute of Technology,
Faculty of Technology & Engineering, CHARUSAT University, India
2
Smt.Kundanben Dinsha Patel Department of Information Technology, Chandubhai S Patel Institute of Technology,
Faculty of Technology& Engineering, CHARUSAT University, India
Article Info ABSTRACT
Article history:
Received Feb 26, 2019
Revised Aug 5, 2019
Accepted Aug 29, 2019
Artificial intelligence (AI) is additionally serving to a brand new breed of
corporations disrupt industries from restorative examination to horticulture.
Computers can’t nevertheless replace humans, however, they will work
superbly taking care of the everyday tangle of our lives. The era is
reconstructing big business and has been on the rise in recent years which has
grounded with the success of deep learning (DL). Cyber-security, Auto and
health industry are three industries innovating with AI and DL technologies
and also Banking, retail, finance, robotics, manufacturing. The healthcare
industry is one of the earliest adopters of AI and DL. DL accomplishing
exceptional dimensions levels of accurateness to the point where DL
algorithms can outperform humans at classifying videos & images. The
major drivers that caused the breakthrough of deep neural networks are the
provision of giant amounts of coaching information, powerful machine
infrastructure, and advances in academia. DL is heavily employed in each
academe to review intelligence and within the trade-in building intelligent
systems to help humans in varied tasks. Thereby DL systems begin to crush
not solely classical ways, but additionally, human benchmarks in numerous
tasks like image classification, action detection, natural language processing,
signal process, and linguistic communication process.
Keywords:
Algorithms
Applications
Artificial intelligence
Deep learning
History
Machine learning
Models
Neural network
Neural network rules
Revolution
Copyright © 2020 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Priyanka Patel,
U & P U Patel Department of Computer Engineering, Chandubhai S Patel Institute of Technology,
CHARUSAT University,
Changa, 388421, Gujarat, India.
Email: priyankapatel.it@charusat.ac.in
1. INTRODUCTION
An interesting aspect regarding neural systems is to what extent they have taken to be an overnight
achievement. Andrey Kurenkove illustrate a very nice history chart in Figure 1. History returns the whole
distance to the 1940s. Deep learning has extremely just taken off in the last five years from the journey of
Neural Network. The reason is the expanded accessibility of name information alongside the significantly
expanded computational throughput of current processors.
Deep learning algorithms are composed of algorithms which allows software to instruct and train
itself to perform tasks adeptly with the huge dataset of content or text, images, sounds, speeches, videos or
time series frames by introducing with multi-layered neural networks with huge amounts of data. Deep
learning (DL) is also known like it is the subset of machine learning (ML) made out of calculations that allow
programming to prepare itself to perform everyday jobs, similar to discourse and picture acknowledgment,
by uncovering multi-layered neural systems to tremendous measures of information.
Artificial Intelligence (AI) could be a machine that solves your real-life problems while not
obtaining any humans concerned in it, or to be specific we are able to say that a package that is sort of a little
kid that wants some instruction and training and afterward it's capable of resolution your real-life issues. An

Int J Elec & Comp Eng ISSN: 2088-8708 
The upsurge of deep learning for computer vision applications (Priyanka Patel)
539
assessment of the impact of AI requires a comprehension of both the practice of innovation technology and
law. While the referenced examinations assessed exercises of attorneys and current innovations, they don't
adequately connect with future conditions of innovation and if you would like a particular definition by Alan
Turing, AI defined as "the science of making computers do things that require intelligence when done by
humans. Now, Artificial Intelligence could be a big tree that has several branches to review and specialize
with it" [1, 2].
Figure 1. Brief history of neural network and deep learning
Figure 2 is brief of Artificial Intelligence tree and difference between artificial intelligence, Machine
Learning, Neural Network, Natural Language Processing, and Deep Learning:
AI: Building systems Building frameworks that can do savvy things implies PCs with the capacity to reason
like humans.
ML: It is likewise a subcategory of Artificial Intelligence. Structure a frameworks which will gain as a
matter of fact implies machine with the capacity to learn without being unequivocally customized.
NLP: It is a subcategory of Artificial Intelligence. Structure a frameworks that can comprehend language. It
is a subcategory of AI.
DL: It is a subcategory of Machine Learning. Structure frameworks that utilization DNN on a substantial
arrangement of information implies computers with the ability to learn by using artificial neural networks,
which were roused by the model and capacity of the model cerebrum as shwon in Figure 2.
NN: A biologically enlivened network of Artificial Neurons. Artificial Intelligence is encompassed with
numerous diverse sub-specialties. Extensively, they can be gathered: Reasoning tools-to generate conclusion
from available data or knowledge like picture, content, videos. Unsupervised deep learning tools- create
general systems that can be trained with tiny amount of data. The main objective of this kind of learning
research is to pre-train a model like discriminator or encoder network to be used for other tasks. Supervised
tools-To understand comprehend importance and setting through classification and regression. Need to guide
to teach the algorithm what conclusions it should come up with. Learning tools- To put on and broaden
understanding through machine learning and predictive analysis. Optimization tools- To put on information
to express inquiries through expert frameworks.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 10, No. 1, February 2020 : 538 - 548
540
Figure 2. Type of learning and its applications structure systems that use dnn on a large set of data [3, 4]
2. MACHINE LEARNING APPROACH
ML is a kind of artificial brainpower which allows software applications to conclusion with
progressively precise in expecting results without being expressly revised. The essential start of ML is to
construct calculations which will get input data and utilize measurable examination to predict associate
output value within an adequate range. Figure 3 shows the straightforward process of machine learning. In
machine learning features are manually extracted and the model creates after extraction of features. These
approach follows shallow network as shown in Figure 4. We can improve network through providing more
examples and more training data to the network. In deep learning, the features from the dataset are
automatically extracted. The Performs of the model is ―end-to-end learning‖ means deep network, not like a
shallow network. We can improve the network by providing huger dataset so, a deep learning algorithm is a
scale with data. Figure 5 is the example of DL.
Figure 3. Basic machine learning
process flow
Figure 4. Machine learning approach to hand craft features
and classification
Figure 5. Deep learning approach to auto extracting features and classification [5]

541
3. NEURAL NETWORKS (NN) AND DEEP LEARNING
NN and DL presently give the most effective solutions to several issues in image recognition,
speech recognition, text recognition, time series problems, video processing, and natural language process.
Deep learning is nothing but an algorithm which is way smarter than shallow machine learning algorithms
but it is not an easy task. Deep neural networks are harder to train and required graphical processing support
for better and fast results. Nowadays deep learning applications are used everywhere many examples of deep
learning are available in current times such as Google’s self-driven car, apples face recognition system, SIRI,
Cortana personal assistant all those are developed using deep learning algorithm also newly introduced
Amazon Go stores also includes deep learning. Three Capabilities of Deep Learning first is
Generalizability — how precisely the machine will execute estimation on given data which has not been
formulated yet? Second is Trainability how rapidly decilitre a DL framework will get accustomed to its
concern. And third is Expressibility — this feature delineates how well a machine will evaluate general
estimations. Deep learning, for instance, interpretability, measured quality, transferability, inertness, ill-
disposed soundness, and security.
4. CURRENT CHALLENGES
In current time Deep Learning has turned out to be one in everything about first investigation
regions in creating astute machines. The majority of notable applications like Image and Speech Recognition,
Text processing, and NPL-Natural language processing of AI are driven by Deep Learning nowadays very
easily. Deep Learning algorithms copy human cerebrums exploitation artificial neural networks and more and
increasingly more profoundly figure out how to precisely take care of a given issue. However there are vital
challenges in Deep Learning systems that we've got to seem out for lots and lots of information, a large
amount of data and plenty of information, Overfitting in neural networks, Hyperparameter Optimization,
Requires high-performance hardware, Neural networks are in essence a Black-box, Lack of Flexibility and
Multitasking.
Deep learning is a methodology that models human theoretical reasoning or possibly speaks to
endeavor to approach it rather utilizing. Though, this innovation has an also some set of disadvantages too.
Management of Constant Input Data: In DL, a training process depends on examining a lot of information.
Albeit, speedily flowing input data and give a little period of time toward guaranteeing a productive training
process. Which is the main cause that data researchers need to adjust their algorithm in the manner in which
neural networks will deal with a lot of constant input data? Guaranteeing Conclusion Transparency: One
more critical disadvantage of deep learning programming is, DL is unequipped for furnished opinions
because it has already achieved a convincing conclusion. Unlike in case of ancient machine learning, we can't
track an associate algorithm formula to search why, our system has fixed that it's a cat on an image, not a
tiger. To address blunders in deep learning program or algorithm, we need to update the entire calculation
and algorithm too. Resource-Demanding Technology: It is an altogether resource-demanding technology.
It needs additional powerful high-performance GPUs, means its work on CPUs but it gives tremendous
results on GPUs, it also required large amounts of storage space to training a model, etc. Moreover, the
technology takes a longer training time as compared to ancient ML. Despite all its challenges, deep learning
explorers advanced techniques for those researchers who have the intention to use unstructured big data
analytics. Indeed, DL tasks of data processing give significant benefits to Educators, analyst, corporations.
5. RECENT DEVELOPMENTS
There are so many computer vision problems and issues like classification reorganization,
identification, language processing, video processing, gesture detection, robotics etc. will currently in
the world of progressing deep learning area all Computer vision problems be measured as resolved. In recent
model-based development, models based on CNN- Convolutional Neural Networks have revolutionized
the entire field of computer vision. It is now very easy to develop through several Deep Learning
configurations utilizing adjusting with pre-trained weights. For example, classification through ImageNet.
Nevertheless, the more security issues of object detection and segmentation need advanced ways to crack the
solution. Object detection consists of learning the objects and drawing a rectangle bounding box, whereas
segmentation targets to spot precise pixels that fit every object. One amongst most variations through image
classification is that a similar image could contain many objects and people might remain in totally diverse
sizes, lighting, proportion, and partly occluded. Professor Nick Reed, academy director at the Transport
Research Laboratory (TRL), agrees that deep learning is a very important tool, but one that raises serious
concerns. "Deep learning is somewhat opaque. It can be hard to understand the rules or knowledge learned by
the system. In the case of self-driving cars, this may become important in the event of a collision. For

 ISSN: 2088-8708
542
example, if a vehicle is dependent on a deep learning algorithm, it may be difficult to understand how the
vehicle used the available information to determine its actions, subsequently resulting in a lack of clarity over
liability."
6. TOWARDS DEEP LEARNING
Deep learning a paradigm of Machine learning which has shown incredible promises in recent time.
This is because of the fact that Deep Learning shows a great analogy with the functioning of the human brain.
In below Figure 6 shows the Traditional Machine Learning approaches. It seems in a picture of machine
learning that features extraction and classification are done separately and both concerned with the complex
design and lots of significant mathematics, after doing it very precisely sometimes it was not even efficient,
and didn’t perform well means in concern with real-world applications the accuracy level wasn’t suitable.
Figure 6. Traditional machine learning flow
7. AT PRESENT, ABOUT DEEP LEARNING
DL networks will perform feature extraction and classification in one shot, that’s why sometimes
people use to call one short learning instead of deep learning. Which implies it just need to plan and design
one model. The fundamental advantage of deep learning over ML algorithm is, DL has the ability to create
new features from partial series of features located in the training dataset. Also as seen in Figure 7 Feature
Extraction and classification will be done in one short. DL model has the convenience of GPUs and GPU
will process huge amounts of labeled data in parallel at high speeds empowers. DL models to be much faster
than ML methods. In DL Model with backpropagation algorithm, effective loss function like ReLu, and a big
amount of parameters, these DL networks are capable to learn extremely complex and abstract features from
the set, traditional ML need to handcraft the features and no abstract features are there. Some Pytorch,
TensorFlow, and Keras are high-level open source libraries are there with DL frameworks for easy
implementation. Nowadays the newest trend in research and development in Machine Learning is Deep
learning, why because DL methods have carried ground-breaking advances in computer vision and machine
learning. Deep Learning is open, outperform, state-of-art technology. In recent time many new application
like speech, video, robotics, security makes practically feasible.
Figure 7. Deep learning flow
8. TYPE OF ARCHITECTURES
Researchers have created a variety of architectures which we can see in below chart that those
existed from the late 1940s, architectures and approaches are modified in day by day development and new
architectures are created but after 2006 the new architectures and graphical processing units (GPUs) brought
them to the pole of AI. From the last two decades, deep learning architectures have greatly expanded
the problems neural networks can address. Deep learning is much more than the neural network follows
the Figure 8 introduce from many research papers by Favio Vázquez.
Table 1 shows the Neural network to Deep learning models, rules and some functions with their
applications. McCulloch–Pitts model, Hebb Rule, Perceptron, Adaptive linear element, Multi-Layer
Perceptron (MLP), Hopfield Network Circuit, Back-Propagation, Support vector Machine (SVM),
Boltzmann Machine, Restricted Boltzmann Machine (RBM), Deep Boltzmann Machine (DBM), Auto
encoder, Recurrent Neural network (RNN), Convolution neural Network or CNN or ConvNet, Long short-
term memory (LSTM), Deep belief network (DMN), Deep Auto-encoder, Sparse Auto encoder, Attention
based LSTM (ALSTM), Generative Adversarial Network (GAN).

543
Figure 8. Artificial intelligent, machine learning, deep learning approach's revolution time line
Table 1. Models and rule with their graphical representation and applications
Proposed models Graphical representation Applications Timeline
McCulloch–Pitts
model
- First ever mathematical
model
- Linear threshold gate
- Weights and bias are
fixed
- Generates binary output
1943 [6]
Hebb Rule
- It is a learning Rule.
- It is used to specify the
weight in proportion to
activation function.
- It works well if the
input patterns are
orthogonal or
uncorrelated.
1949 [7]
Perceptron
- Used for supervised
learning of binary
classifiers
- Classification and
- Regression
1958 [8]
Adaptive linear
element
- Single layer ANN
- Adaptive linear unit
/neuron
1960 [9, 10]
Multi-Layer
perceptron
- Feedforward artificial
neural network.
- Speech Recognition
- Image Recognition
- Machine Translation
- Classification
1969
[11, 12]

 ISSN: 2088-8708
544
Table 1. Models and rule with their graphical representation and applications (continue)
Hopfield
Network Circuit
- Recurrent Artificial
Neural Network
- Recognition
1982 [13]
Back-Propagation
- Classification
- Time-series Prediction
- Function Approximation
1970 [14]
Support vector
Machine
- Bioinformatics
- Generalized Predictive
Control
- Object Detection
- Handwriting Recognition
- Text And Hypertext
Categorization
- Images Classification
- Protein Fold And Remote
Homology Detection
1963 [15]
1992 [16]
Boltzman
Machine
- Pattern recognition
- Combinatorial
optimization
- Speech recognition
1986 [17]
Restricted
Boltzman
Machine
&
Deep Boltzman
Machine
- Pattern Recognition
- Classification
- Combinatorial
Optimization
1986 [18]
Auto encoder
- Used for unsupervised
learning and
dimensionality reduction
- It’s for representation
learning and deep learning
- Image generation
- Data visualization
- Natural language
processing
1986 [19]

545
Recurrent Neural
network
1986
[20-23]
Convolution
neural Network
or
CNN
or
ConvNet
- Image And Video
Recognition
- Image Classification
- Video Analysis
- Natural Language
Processing(NLP)
- Medical Image Analysis
1983 [24]
1999 [25]
Long short-term
memory
- Gesture Recognition
- NLP &Text Compression
- Image Captioning
1997 [26]
Deep belief
network
- Natural Language
Understanding
- Image Recognition
- Failure Prediction
- Information Retrieval
2007
2009 [27]
-
Deep Auto
encoder
learning and
- It’s for representation
learning and deep learning
- Image generation
- Data visualization
- Natural language
processing
2000s [28]
-

 ISSN: 2088-8708
546
Sparse Auto
encoder
learning and
- Object Recognition
- Medical Diagnosis
2000s
[29-31]
Attention based
LSTM
- Used to predict target
word
- Deep NLP
2010s
[32-34]
Generative
Adversarial
Network
- unsupervised machine
learning
- Reconstruct 3D models
from image.
2014
[35-37]
9. HOW DEEP LEARNING WORKS
Deep learning working and the neural network working is almost the same. DL works for both
supervised learning and unsupervised learning. And as earlier discussed in paper DL can solve many
complex problems of computer vision which were not easily solved by Machine learning. It made up of two
principal stages: training and Testing.
a) The training stages - labeling process of large amounts of data, and define the identical features. The
training model compares these features and keeps them to make precise reasoning and conclusion when it
encounters related data next time. The DL training process encompasses the following stages:
1) ANNs explore a set of binary false/true questions or.
2) Extract numeral values from data bar.
3) Classify data correspond to the proper responses got.
4) Labelling Data.
b) The testing stages - label new unexposed data using their previous knowledge and then make a
conclusion. In ancient machine learning, new feature extraction will be extracted from a fundamental data
stacked into the machine. Analyst formulates ML instructions and corrects the errors encountered by a
machine.
This methodology wipes out a negative overtraining impact much of the time be seen in deep learning.
In ML, the analyst supplies both examples and training data to assist the system to make correct decisions
by the machine this norm is known as supervised learning. In other words, in an ancient machine
learning, a computer solves a huge number of tasks, however, it can't mount such undertakings without
human control. Diversity between machine learning and deep learning:
1) Deep learning requires unlabelled training data in huge quantity to make concise conclusions while
Machine Learning will utilize a small amount of data from the analyst.
2) Machine learning expects features to be precisely recognized by analyst while deep learning generates
new features independently.
3) Unlike machine learning, deep learning needs GPU-high-performance hardware.

547
4) Machine learning isolates undertakings into little parts and afterward join and conclude into one
output while deep learning solves the problem on the end-to-end basis.
5) In comparison with Machine learning, deep learning needs considerably more time to train.
6) Machine learning gives enough transparency for its decisions than deep learning.
The idea of deep learning infers that the machine makes its usefulness independent decisions from
anyone else as long as it is conceivable at the present time. To surmise, applications of deep learning utilize
a various levelled methodology including deciding the most imperative attributes to look at.
10. CONCLUSION
An observations of this paper is Unsupervised learning had been a major effect on in reviving deep
learning, but the supervised learning has great success in the 1960s to 2000s so, deep learning has since been
overshadowed, but then again after 2000s deep learning is on top to solve supervised and unsupervised
complex problems. Although we have not focused to review these kinds of learning. The future expectation
for deep learning to become far more important in the longer term. Speech recognition, natural language
processing, and video analysis are largely unsupervised: we reviewed the different models of deep learning to
solve these kinds of problems. Deep learning algorithms and models are trained end-to-end and that uses
reinforcement learning to decide where to look. Deep learning and reinforcement learning already outperform
in one short classification and feature extraction and produce impressive results. Natural language processing
is another area in which deep learning is doing large impact of experiments over the next few years. Speech
recognition and video analysis are also major complex issues in computer vision. For example NLP problem,
we expect
a model which use Recurrent Neural Network to understand documents content or sentences will become
the better rule for selectively attending to one part at a time. Eventually, major progress in artificial
intelligence will come about through systems deep learning. In spite of the fact that to produce good result
deep learning has been used for image, speech, video and handwriting recognition.
REFERENCES
[1] H. A. Simon, ―Artificial intelligence: an empirical science,‖ Artificial Intelligence, vol. 77, pp. 95-127, 1995.
[2] T. Dettmers, T. Dettmers, T. Dettmers, E. Shelhamer and T. Dettmers, "Deep Learning in a Nutshell: History and
Training | NVIDIA Developer Blog", NVIDIA Developer Blog, 2019. [Online]. Available:
https://devblogs.nvidia.com/deep-learning-nutshell-history-training.
[3] https://www.xenonstack.com/blog/data-science/overview-of-artificial-intelligence-and-role-of-natural-language-
processing-in-big-data
[4] https://i2.wp.com/vincejeffs.com/wp-content/uploads/2017/03/AI_Automated_Intelligence.png
[5] https://www.mathworks.com/discovery/deep-learning.html.
[6] W. S. McCulloch and W. Pitts, ―A logical calculus of the ideas immanent in nervous activity,‖ The bulletin of
mathematical biophysics, vol. 5, pp. 115-133, 1943.
[7] D. O. Hebb, ―The organization of behavior. A neuropsychological theory,‖ 1949.
[8] F. Rosenblatt, ―The perceptron: a probabilistic model for information storage and organization in the brain,‖
Psychological review, vol. 65, pp. 386, 1958.
[9] B. Widrow, ―Adaptive adaline Neuron Using Chemical,‖ memistors, 1960.
[10] Eriksson, K., & Johnson, C. (1988). An adaptive finite element method for linear elliptic problems. Mathematics of
Computation, 50(182), 361-383.
[11] M. L. Minski and S. A. Papert, ―Perceptrons: an introduction to computational geometry,‖ MA, MIT Press,
Cambridge, 1969.
[12] Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1988). Learning representations by back-propagating
errors. Cognitive modeling, 5(3), 1.
[13] J. J. Hopfield, ―Neural networks and physical systems with emergent collective computational abilities,‖
Proceedings of the national academy of sciences, vol. 79, pp. 2554-2558, 1982.
[14] D. E. Rumelhart, et al., ―Learning representations by back-propagating errors,‖ nature, vol. 323, pp. 533, 1986.
[15] Vapnik and Lerner, ―Introduce the Generalized Portrait algorithm (the algorithm implemented by support vector
machines is a nonlinear generalization of the Generalized Portrait algorithm),‖ 1963.
[16] V. Vapnik and A. Lerner, ―Pattern recognition using generalized portrait method,‖ Automation and Remote
Control, vol. 24, pp. 774-780, 1963.
[17] D. H. Ackley, et al., ―A learning algorithm for Boltzmann machines,‖ Cognitive science, vol. 9, pp. 147-169, 1985.
[18] P. Smolensky, ―Information processing in dynamical systems: Foundations of harmony theory,‖ Colorado Univ at
Boulder Dept of Computer Science, 1986.
[19] D. E. Rumelhart, et al., ―Learning internal representations by error propagation,‖ in Rumelhart D., et al., ―Parallel
distributed processing: Explorations in the microstructure of cognition,‖ Foundations, vol. 1, 1986.
[20] R. J. Williams, et al., ―Learning representations by back-propagating errors,‖ Nature, vol. 323, pp. 533-536, 1986.

 ISSN: 2088-8708
548
[21] K. Fukushima and S. Miyake, ―Neocognitron: A self-organizing neural network model for a mechanism of visual
pattern recognition,‖ Competition and cooperation in neural nets, Springer, Berlin, Heidelberg, pp. 267-285, 1982.
[22] Liang, M., & Hu, X. (2015). Recurrent convolutional neural network for object recognition. In Proceedings of the
IEEE conference on computer vision and pattern recognition (pp. 3367-3375).
[23] Lange, S., & Riedmiller, M. (2010, July). Deep auto-encoder neural networks in reinforcement learning. In The
2010 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
[24] K. Fukushima, S. Miyake and T. Ito, "Neocognitron: A neural network model for a mechanism of visual pattern
recognition", IEEE Transactions on Systems, Man, and Cybernetics, vol. -13, no. 5, pp. 826-834, 1983. Available:
10.1109/tsmc.1983.6313076.
[25] Y. LeCun, et al., ―Object recognition with gradient-based learning,‖ Shape, contour and grouping in computer
vision, Springer, Berlin, Heidelberg, pp. 319-345, 1999.
[26] S. Hochreiter and J. Schmidhuber, ―Long short-term memory,‖ Neural computation, vol. 9, pp. 1735-1780, 1997.
[27] G. E. Hinton, ―Deep belief networks,‖ Scholarpedia, vol. 4, pp. 5947, 2009.
[28] Y. Bengio and Y. LeCun, ―Scaling learning algorithms towards AI,‖ Large-scale kernel machines, vol. 34, pp. 1-
41, 2007.
[29] Ng, A. (2011). Sparse autoencoder. CS294A Lecture notes, 72(2011), 1-19.
[30] C. Doersch, ―Tutorial on variational autoencoders,‖ arXiv preprint arXiv:1606.05908, 2016.
[31] Sun, W., Shao, S., Zhao, R., Yan, R., Zhang, X., & Chen, X. (2016). A sparse auto-encoder-based deep neural
network approach for induction motor faults classification. Measurement, 89, 171-178.
[32] Y. Qin, et al., ―A dual-stage attention-based recurrent neural network for time series prediction,‖ arXiv preprint
arXiv:1704.02971, 2017.
[33] Wang, Y., Huang, M., Zhu, X., & Zhao, L. (2016, November). Attention-based LSTM for aspect-level sentiment
classification. In Proceedings of the 2016 conference on empirical methods in natural language processing (pp.
606-615).
[34] Gao, L., Guo, Z., Zhang, H., Xu, X., & Shen, H. T. (2017). Video captioning with attention-based LSTM and
semantic consistency. IEEE Transactions on Multimedia, 19(9), 2045-2055.
[35] Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., & Bengio, Y. (2014). Generative
adversarial nets. In Advances in neural information processing systems (pp. 2672-2680).
[36] Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V., Radford, A., & Chen, X. (2016). Improved techniques for
training gans. In Advances in neural information processing systems (pp. 2234-2242).
[37] Jahanian, A., Chai, L., & Isola, P. (2019). On the''steerability" of generative adversarial networks. arXiv preprint
arXiv:1907.07171.
BIOGRAPHIES OF AUTHORS
Priyanka P. Patel has received her Bachelor degree in Information Technology from Gujarat
University in the year 2006. She has completed her Master degree in Computer Engineering
from Chandubhai S Patel Institute of Technology, CHARUSAT university in the Year 2013. She
is currently pursuing her PhD degree in computer engineering at CHARUSAT University,
Gujarat, India, and working as assistant professor in Department of Information Technology,
Chandubhai S. Patel of Institute, Gujarat, India. Her research interests include image processing,
Video Analysis, Machine Learning, Deep Learning.
Dr.Amit Thakkar has received his B.E Degree in I.T. from Gujarat University in 2002 and
Master Degree from Dharmsinh Desai University, Gujarat, India in 2007. He has finished his
PhD in the area of Multi Relational Classification from Kadi Sarva Vishwa Vidyalaya (KSV),
Gandhinagar, India in 2016. He is working as an Associate Professor in the Department of
Information Technology in Faculty of Engineering and Technology, CHARUSAT University,
Changa, Gujarat Since 2002. He has published more than 50 research papers in international,
national journals and conferences of repute. He has more than fifteen years of teaching
experience. He has developed subject proficiency in Data Structures, Database Management
System, Data Compression and Data Mining His research interests include database, Data
Mining, Machine Learning, Deep Learning.

The upsurge of deep learning for computer vision applications

More Related Content

What's hot (20)

Similar to The upsurge of deep learning for computer vision applications (20)

More from IJECEIAES (20)

Recently uploaded (20)

The upsurge of deep learning for computer vision applications