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

net/publication/374023047

KERAS DEEP LEARNING PACKAGE IN PYTHON: A REVIEW

Article · September 2023

CITATION READS

1 1,002

4 authors, including:

Dalia shihab Ahmed Nadia Mahmood Hussien

Mustansiriyah University Mustansiriyah University
11 PUBLICATIONS 11 CITATIONS 103 PUBLICATIONS 224 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Dalia shihab Ahmed on 19 September 2023.

The user has requested enhancement of the downloaded file.

 European Journal of Interdisciplinary Research and Development
Volume-19 September 2023
Website: www.ejird.journalspark.org ISSN (E): 2720-5746
KERAS DEEP LEARNING PACKAGE IN PYTHON: A REVIEW
Dalia Shihab Ahmed1,
Rasha shaker ibrahim Al-badri2,
Firas Ali Hashim3,
Nadia Mahmood Hussien4
1,2,3,4 Computer Science Department Mustansiriyah University Baghdad, Iraq
[email protected]
[email protected]
[email protected]
[email protected]

Abstract:
Python's popularity is growing. It is a member of the languages that are used most frequently,
having supplanted a lot of others in the in the workplace. Python is well-known among
developers for a lot of reasons, one of which is because it offers an exceptionally wide groups
of libraries that people can utilize. This document gives an up-to-date assessment of Keras, an
Application Programming Interface (API) for deep learning built on the Tensor Flow machine
learning framework and developed using Python. The mentioned package utilizes Tensor Flow,
and others, to integrate deep learning models such as neural network graphs for identifying
medical problems, and others. The purpose of this paper is to guide readers to Keras, a library
which offers exceptionally strong and abstract building blocks for deep learning networks.

Keywords: Keras, PyTorch, Tensor Flow, CNN, neural network graphs, Deep learning.

Introduction
A rise of machine learning techniques over the past few decades has resulted in widespread use
in a variety of industries. Significant increases in terms of training set quantity and available
resources for processing have led in the development of a new generation of applications. As
the outcome of this advancement, the use and spread of deep learning has grown[1][2]. There
are many languages of programming provide packages and tools for populating neural networks
that are artificial. Yet, Python has emerged as the clear winner in this arena[3]. Python is a
straightforward and easy-to-use computing language with several scientific packages such as
Num, Scipy, Keras, and others for performing particular simulations, regression analysis, and
determining basic differential equations[4]. The field of deep learning is a subfield of Machine
Learning that employs artificial neural networks to solve issues with big datasets. Additionally,
the technology is influenced by the functioning of the human brain. Furthermore, without human
interaction, artificial neuron networks evaluate enormous data sets to uncover underlying
patterns[5].
Deep Learning approaches gives a solid foundation for Supervised Learning. The learning model
presents greater accuracy of labeled image data by adding more layers. The capabilities of
hardware expanded for computer processing, the rising size of data needed for training the model,
and the developing advancement in research to ML algorithms and information handling are
24 | P a g e
 European Journal of Interdisciplinary Research and Development
Volume-19 September 2023
Website: www.ejird.journalspark.org ISSN (E): 2720-5746
three key causes driving the growing acceptance of deep learning today. Deep learning
applications include, but are not limited to, machine vision, recognition of speech, natural
language processing, and other associated fields. [6][7]. On top of the Tensor Flow machine
learning framework, Keras is a Python-based deep neural network API. It was designed to allow
for speedy experimentation while also providing a delightful development experience. Keras's
goal is to provide every one that wants to have an unfair advantage in shipping ML-powered
apps[8]. This paper is organized as follows: First, we introduce Python, its packages and deep
learning. Second, we will describe the theoretical concepts which represents in related works.
Third, we show what is a keras with its goals, advantages and disadvantage. Finally, we offer a
conclusion for this paper.

1. Related work
In [9].This work investigates the problem of Arabic script identification in the context of
OCRing-based recognition. Utilizing Keras and Tensor Flow, an innovative design is suggested
that combines the Rapid Gradient Sign Method (FGSM) with adversarial image creation
throughout the training phase. Furthermore, the research uses deep learning to improve the
OCRing accuracy of picture improvement, position, layout analysis, and recognition in
multilingual systems. A specially trained deep learning algorithm that uses bounding box
regression trained using Keras and Tensor Flow will be used to detect RoIs. Using the
reinforcement strategy, this subject studies a modification of the Page Segmentation strategy
(PSM) to enhance OCRing parameters and the precision of an Arabic OCRing system.
In[10].The first stage of this method is Region of Interest Segmentation, which is based on the
color space segmentation methodology and employs a pre-set color range to separate pixels
(hand) from the background (pixels not in the targeted area of interest). The images with
segments are then sent into a Convolutional CNN model for categorization of images in the
system's second phase. For picture training, we used the Python Keras library. The system
highlighted the significance of picture segmentation in the recognition of hand gestures. The best
model has an efficiency of 58 percent, meaning it is approximately 10% more than the precision
obtained without image segmentation.
This article [11]. The TSFEDL library is shown. It brings together 22 cutting-edge methods
enabling time series feature extraction and estimation, using convolution and recurring deep
neural networks for use in a wide range of data mining situations. Tensor Flow + Keras and
PyTorch modules are used in the library, which is licensed via the AGPLv3 license. The
efficiency of the assessment of the designs given in the suggestion validates the utility of the
Python module.
In[12]. They suggest AI that identifies facial expressions by utilizing the multiple layers of CNN.
Deep Face Emotion Recognizer (FER) evaluation, which includes datasets and methodologies
to provide light regarding these fundamental difficulties. Initially the FER plan for obtaining
advice at each level is implemented, which includes necessary historical data. The FER challenge
dataset from kaggle was used for the experiment. The development context includes the Python
libraries keras, Tensor Flow, and cv2.

25 | P a g e
 European Journal of Interdisciplinary Research and Development
Volume-19 September 2023
Website: www.ejird.journalspark.org ISSN (E): 2720-5746
In [13]. Their goal in implementation this system is to research and evaluate (ML) methods for
recognition and identification of individuals sporting fa Masks in pre-recorded movies is their
project's aim., images, or real-time (real-time) based on Keras. The suggested approach employs
PCA, or principal component analysis, and using HAAR Cascade Algorithms.
In [14]. SciANN is presented as a Python instrument for deep learning with artificial neural
networks based on physics and computational science. SciANN builds deep neural network and
algorithms for optimization using the well-known deep learning packages Tensor Flow and
Keras, gaining features from Keras like batch optimization and modeling repetition for learning
by transfer. They demonstrate ways the structure can be utilized for fitting curves on separate
information, as well as resolving and finding PDEs in both weak and strong forms, through a
number of cases.
In [15]. They propose a Tensor Flow-Keras model implementation of graph convolution and
graph pooling layers that allows for Integration that is both seamless and adaptable into normal
Keras layers to put up graph in a functional manner. They created the Keras Graph CNN, which
is built based Tensor Flow and Keras and focuses on a transparent tensor structure transmitted
among layers as well as user-friendliness.
In[16]. This study formalizes the challenge by augmenting the well-known Keras Python
framework with the concept of Composed Neural Networks (CpNNs). In Keras, CpNNs
formalize the simultaneous building of many interacting neural networks. Furthermore, They
enable modular compilation from a given CpNN to C code utilizing the proposed semantics. The
resulting code can be used to do Worst-Case Execution Time (WCET) analysis.
For the comparisons they provided, their method exceeds Esterel with a median WCET reducti
on of 64.06% and Tensor Flow Lite with a median measurable WCET reduction of 62.08%.
In [17]. They began by providing a quick introduction to the field of deep learning (DL) before
going on to evaluate the prominent options for scalable platforms for DL research. They
examined the Keras ecosystem thoroughly for insight into what makes it distinctive, as well as
sample code to determine how straightforward the framework is to use for building DL models.
They produce digital elevations information for a segmentation framework using the Keras
package, instead of standard human processing processes such as tiling rasters to samples of
masking samples out the study's area, and applying digital elevation model dependents.
In [19]. Regarding distributed neural network learning over numerous GPUs or CPUs, they offer
a compact Python platform. The well-known Keras machine learning library. To coordinate the
training process, the (MPI) protocol is employed, and the system is well suitable for task give at
supercomputing locations

2. Keras
Keras is a free more advanced neural network toolkit written in the Python which may be used
on Theano, Tensor Flow, or CNTK. François Chollet, a Google developer, designed it. It was
created to be user-friendly, extendable, and modular for the purpose to allow for faster
experimenting with deep neural networks. It not only supports Convolutional and Recurring
Network alone, as well as in combined. Keras gives explicit feedback when an error occurs,
reducing the amount of user activities for the vast majority of frequent use cases.: Simple to learn

26 | P a g e
 European Journal of Interdisciplinary Research and Development
Volume-19 September 2023
Website: www.ejird.journalspark.org ISSN (E): 2720-5746
and apply, Extremely adaptable.Keras gives all of its programmers a lot of freedom through the
integration of low-level machine learning languages like Tensor Flow or Theano, which means
that everything written in the language of origin may be executed by Keras [5][8][20][21][22].
Keras is intended to lessen cognitive strain by accomplishing the following objectives:
● Provide simple, consistent user interfaces.
● Reduce the amount of actions required for typical use cases.
● Give clear and relevant messages of error.
● Follow the principle of gradual complexity disclosure: It's simple to get started, and you
may acquire as you go to finish complicated workflows.
● Assist you in writing clear, readable code

3.1 Keras benefits include the following:

● The speedier deployment of network models is simple to grasp and incorporate.
● It has a large market community support because most AI businesses want to use it. It
supports several backends, which means you can use either of Tensor Flow, CNTK, or
Theano with Keras as a backend depending on your needs.
● It also supports cross-platform due to its ease of deployment.
Keras may be installed on iOS and CoreML, Tensor Flow on Android Android Web browser
with.js support Cloud engin and the Raspberry Pi
● It allows data parallelism, which means Keras may be trained simultaneously on several
GPUs.
● It allows information parallelism, which implies that Keras may be taught on several GPUs
at the same time, shortening train time and interpreting large amounts of data.

3.2 Keras' disadvantages

The only drawback is that Keras has a number of pre-configured layers and will not allow you
to establish an abstraction layer because it is unable to handle minimal APIs. Only high-level
APIs executing on front of the back processor (Tensor Flow, Theano, and CNTK) are supported.

4. Conclusion
With the fast growth of machine learning in numerous domains, large corporations and research
groups have created autonomous and one-of-a-kind tools. Deep Learning with Python teaches
thousands of readers how to fully utilize deep learning's possibilities.Current deep learning
breakthroughs provide interesting new software capabilities such as automatic language
translation, image recognition, and more. Deep learning is increasingly becoming required
expertise for every software developer, and new technologies such as Keras and Tensor Flow are
making it possible. Keras is a Python-based free higher levels neural network package that can
operate on Theano, Tensor Flow, or CNTK. To make deep neural network programming
simpler, Keras offers numerous implementations of frequently used neural network building
blocks like layers, objectives, activation functions, optimizers, and a plethora of tools for
working with image and text data.

27 | P a g e
 European Journal of Interdisciplinary Research and Development
Volume-19 September 2023
Website: www.ejird.journalspark.org ISSN (E): 2720-5746
Acknowledgment
The Author would like to thank Mustansiriyah University (https://uomustansiriyah.edu.iq/)
Baghdad –Iraq for its support in the present work.

References
1. Vasilev, I., Slater, D., Spacagna, G., Roelants, P., & Zocca, V. (2019). Python Deep
Learning: Exploring deep learning techniques and neural network architectures with
Pytorch, Keras, and Tensor Flow. Packt Publishing Ltd.
2. Gulli, A., Kapoor, A., & Pal, S. (2019). Deep learning with Tensor Flow 2 and Keras:
regression, ConvNets, GANs, RNNs, NLP, and more with Tensor Flow 2 and the Keras
API. Packt Publishing Ltd.
3. Atienza, R. (2020). Advanced Deep Learning with Tensor Flow 2 and Keras: Apply DL,
GANs, VAEs, deep RL, unsupervised learning, object detection and segmentation, and
more. Packt Publishing Ltd.
4. Hassan, G. M., Hussien, N. M., & Mohialden, Y. M. (2023). Python TCP/IP libraries: A
Review. International Journal Papier Advance and Scientific Review, 4(2), 10-15.
5. Ashraf, M., Ahmad, S. M., Ganai, N. A., Shah, R. A., Zaman, M., Khan, S. A., & Shah, A.
A. (2021). Prediction of cardiovascular disease through cutting-edge deep learning
technologies: an empirical study based on TENSOR FLOW, PYTORCH and KERAS.
In International Conference on Innovative Computing and Communications: Proceedings of
ICICC 2020, Volume 1 (pp. 239-255). Springer Singapore.
6. K. Chauhan dan S. Ram, ―Image Classification with Deep Learning and Comparison
between Different Convolutional Neural Network Structures using Tensor Flow and Keras,‖
Int. J. Adv. Eng. Res. Dev., vol. 5, no. 02, hal. 533–538, 2018
7. Harjoseputro, Y. (2020). A classification Javanese letters model using a convolutional neural
network with KERAS framework. International Journal of Advanced Computer Science and
Applications, 11(10).
8. keras · PyPI
9. Al-Barhamtoshy, H. M., Jambi, K. M., Rashwan, M. A., & Abdou, S. M. (2023). An Arabic
Manuscript Regions Detection, Recognition and Its Applications for OCRing. Transactions
on Asian and Low-Resource Language Information Processing, 22(1), 1-28.
10. Chang, V., Eniola, R. O., Golightly, L., & Xu, Q. A. (2023). An Exploration into Human–
Computer Interaction: Hand Gesture Recognition Management in a Challenging
Environment. SN Computer Science, 4(5), 441.
11. Aguilera-Martos, I., García-Vico, Á. M., Luengo, J., Damas, S., Melero, F. J., Valle-Alonso,
J. J., & Herrera, F. (2023). TSFEDL: A python library for time series spatio-temporal feature
extraction and prediction using deep learning. Neurocomputing, 517, 223-228.

12. Depuru, S., Nandam, A., Sivanantham, S., Amala, K., Akshaya, V., & Saktivel, M. (2022,
December). Convolutional Neural Network based Human Emotion Recognition System: A
Deep Learning Approach. In 2022 Smart Technologies, Communication and Robotics
(STCR) (pp. 1-4). IEEE.

28 | P a g e
 European Journal of Interdisciplinary Research and Development
Volume-19 September 2023
Website: www.ejird.journalspark.org ISSN (E): 2720-5746
13. Hebbale, S., & Vani, V. (2022). Real time COVID-19 facemask detection using deep
learning. learning, 6(S4), 1446-1462.
14. Haghighat, E., & Juanes, R. (2021). SciANN: A Keras/Tensor Flow wrapper for scientific
computations and physics-informed deep learning using artificial neural
networks. Computer Methods in Applied Mechanics and Engineering, 373, 113552.
15. Reiser, P., Eberhard, A., & Friederich, P. (2021). Graph neural networks in Tensor Flow-
Keras with RaggedTensor representation (kgcnn). Software Impacts, 9, 100095.
16. Yang, X., Roop, P., Pearce, H., & Ro, J. W. (2020, March). A compositional approach using
Keras for neural networks in real-time systems. In 2020 Design, Automation & Test in
Europe Conference & Exhibition (DATE) (pp. 1109-1114). IEEE.
17. Moolayil, J., & Moolayil, J. (2019). An introduction to deep learning and keras. Learn Keras
for Deep Neural Networks: A Fast-Track Approach to Modern Deep Learning with Python,
1-16.
18. Soliman, A., & Terstriep, J. (2019, November). Keras Spatial: Extending deep learning
frameworks for preprocessing and on-the-fly augmentation of geospatial data.
In Proceedings of the 3rd ACM SIGSPATIAL International Workshop on AI for
Geographic Knowledge Discovery (pp. 69-76).
19. Anderson, D., Vlimant, J. R., & Spiropulu, M. (2017). An MPI-based Python framework for
distributed training with Keras. arXiv preprint arXiv:1712.05878.
20. Mohialden, Y. M., Hussien, N. M., Mohammed, S. J., & Mohammed, I. S. (2022). Recent
hybrid machine learning algorithm applications: a review. Journal of Information
Technology and Informatics, 2(1), 1-4.
21. Tan, S. W. B., Naraharisetti, P. K., Chin, S. K. and Lee, L. Y.(2020). Simple Visual-Aided
Automated Titration Using the Python Programming Language,ACS Publications, 2020
22. Ali, J. J., Shati, N. M., & Gaata, M. T. (2020). Abnormal activity detection in surveillance
video scenes. TELKOMNIKA (Telecommunication Computing Electronics and Control),
18(5), 2447-2453.

29 | P a g e

View publication stats