Intelligent Computing Proceedings Of The 2020 Computing Conference Volume 3 1st Ed Kohei Arai

Intelligent Computing Proceedings Of The 2020
Computing Conference Volume 3 1st Ed Kohei Arai
download
https://ebookbell.com/product/intelligent-computing-proceedings-
of-the-2020-computing-conference-volume-3-1st-ed-kohei-
arai-22501384
Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.
Intelligent Computing Proceedings Of The 2020 Computing Conference
Volume 2 1st Ed Kohei Arai
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2020-computing-conference-volume-2-1st-ed-kohei-arai-22501386
Intelligent Computing Proceedings Of The 2020 Computing Conference
Volume 1 1st Ed Kohei Arai
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2020-computing-conference-volume-1-1st-ed-kohei-arai-22501388
Intelligent Computing Proceedings Of The 2021 Computing Conference
Volume 3 Lecture Notes In Networks And Systems 285 1st Ed 2021 Kohei
Arai Editor
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2021-computing-conference-volume-3-lecture-notes-in-networks-and-
systems-285-1st-ed-2021-kohei-arai-editor-34418888
Intelligent Computing Proceedings Of The 2022 Computing Conference
Volume 3 Kohei Arai
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2022-computing-conference-volume-3-kohei-arai-43882172

Intelligent Computing Proceedings Of The 2021 Computing Conference
Volume 1 1st Edition Kohei Arai
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2021-computing-conference-volume-1-1st-edition-kohei-arai-33448520
Intelligent Computing Proceedings Of The 2021 Computing Conference
Volume 2 Lecture Notes In Networks And Systems 284 1st Ed 2021 Kohei
Arai Editor
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2021-computing-conference-volume-2-lecture-notes-in-networks-and-
systems-284-1st-ed-2021-kohei-arai-editor-34418886
Intelligent Computing Proceedings Of The 2022 Computing Conference
Volume 1 Kohei Arai
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2022-computing-conference-volume-1-kohei-arai-43882166
Intelligent Computing Proceedings Of The 2022 Computing Conference
Volume 2 Kohei Arai
https://ebookbell.com/product/intelligent-computing-proceedings-of-
the-2022-computing-conference-volume-2-kohei-arai-43882170
Intelligent Computing And Optimization Proceedings Of The 7th
International Conference On Intelligent Computing And Optimization
2023 Ico2023 Volume 4 1st Edition Pandian Vasant
https://ebookbell.com/product/intelligent-computing-and-optimization-
proceedings-of-the-7th-international-conference-on-intelligent-
computing-and-optimization-2023-ico2023-volume-4-1st-edition-pandian-
vasant-230832184

Advances in Intelligent Systems and Computing1230
Kohei Arai
Supriya Kapoor
Rahul Bhatia Editors
Intelligent
Computing
Proceedings of the 2020 Computing
Conference, Volume 3

Advances in Intelligent Systems and Computing
Volume 1230
Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
Advisory Editors
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
Rafael Bello Perez, Faculty of Mathematics, Physics and Computing,
Universidad Central de Las Villas, Santa Clara, Cuba
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
Hani Hagras, School of Computer Science and Electronic Engineering,
University of Essex, Colchester, UK
LászlóT. Kóczy, Department of Automation, Széchenyi István University,
Gyor, Hungary
Vladik Kreinovich, Department of Computer Science, University of Texas
at El Paso, El Paso, TX, USA
Chin-Teng Lin, Department of Electrical Engineering, National Chiao
Tung University, Hsinchu, Taiwan
Jie Lu, Faculty of Engineering and Information Technology,
University of Technology Sydney, Sydney, NSW, Australia
Patricia Melin, Graduate Program of Computer Science, Tijuana Institute
of Technology, Tijuana, Mexico
Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro,
Rio de Janeiro, Brazil
Ngoc Thanh Nguyen
, Faculty of Computer Science and Management,
Wrocław University of Technology, Wrocław, Poland
Jun Wang, Department of Mechanical and Automation Engineering,
The Chinese University of Hong Kong, Shatin, Hong Kong

The series“Advances in Intelligent Systems and Computing”contains publications
on theory, applications, and design methods of Intelligent Systems and Intelligent
Computing. Virtually all disciplines such as engineering, natural sciences, computer
and information science, ICT, economics, business, e-commerce, environment,
healthcare, life science are covered. The list of topics spans all the areas of modern
intelligent systems and computing such as: computational intelligence, soft comput-
ing including neural networks, fuzzy systems, evolutionary computing and the fusion
of these paradigms, social intelligence, ambient intelligence, computational neuro-
science, artiﬁcial life, virtual worlds and society, cognitive science and systems,
Perception and Vision, DNA and immune based systems, self-organizing and
adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics
including human-machine teaming, knowledge-based paradigms, learning para-
digms, machine ethics, intelligent data analysis, knowledge management, intelligent
agents, intelligent decision making and support, intelligent network security, trust
management, interactive entertainment, Web intelligence and multimedia.
The publications within“Advances in Intelligent Systems and Computing”are
primarily proceedings of important conferences, symposia and congresses. They
cover signiﬁcant recent developments in theﬁeld, both of a foundational and
applicable character. An important characteristic feature of the series is the short
publication time and world-wide distribution. This permits a rapid and broad
dissemination of research results.
** Indexing: The books of this series are submitted to ISI Proceedings,
EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink **
More information about this series athttp://www.springer.com/series/11156

Kohei AraiSupriya Kapoor
Rahul Bhatia
Editors
IntelligentComputing
Proceedings of the 2020 Computing
Conference, Volume 3
123

Editors
Kohei Arai
Faculty of Science and Engineering
Saga University
Saga, Japan
Rahul Bhatia
The Science and Information
(SAI) Organization
Bradford, West Yorkshire, UK
Supriya Kapoor
The Science and Information
(SAI) Organization
Bradford, West Yorkshire, UK
ISSN 2194-5357 ISSN 2194-5365 (electronic)
Advances in Intelligent Systems and Computing
ISBN 978-3-030-52242-1 ISBN 978-3-030-52243-8 (eBook)
https://doi.org/10.1007/978-3-030-52243-8
©Springer Nature Switzerland AG 2020
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microﬁlms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional afﬁliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Editor’s Preface
On behalf of the Committee, we welcome you to the Computing Conference 2020.
The aim of this conference is to give a platform to researchers with fundamental
contributions and to be a premier venue for industry practitioners to share and
report on up-to-the-minute innovations and developments, to summarize the state
of the art and to exchange ideas and advances in all aspects of computer sciences
and its applications.
For this edition of the conference, we received 514 submissions from 50+
countries around the world. These submissions underwent a double-blind peer
review process. Of those 514 submissions, 160 submissions (including 15 posters)
have been selected to be included in this proceedings. The published proceedings
has been divided into three volumes covering a wide range of conference tracks,
such as technology trends, computing, intelligent systems, machine vision, security,
communication, electronics and e-learning to name a few. In addition to the con-
tributed papers, the conference program included inspiring keynote talks. Their
talks were anticipated to pique the interest of the entire computing audience by their
thought-provoking claims which were streamed live during the conferences. Also,
the authors had very professionally presented their research papers which were
viewed by a large international audience online. All this digital content engaged
signiﬁcant contemplation and discussions amongst all participants.
Deep appreciation goes to the keynote speakers for sharing their knowledge and
expertise with us and to all the authors who have spent the time and effort to
contribute signiﬁcantly to this conference. We are also indebted to the Organizing
Committee for their great efforts in ensuring the successful implementation of the
conference. In particular, we would like to thank the Technical Committee for their
constructive and enlightening reviews on the manuscripts in the limited timescale.
We hope that all the participants and the interested readers beneﬁt scientiﬁcally
from this book andﬁnd it stimulating in the process. We are pleased to present the
proceedings of this conference as its published record.
v

Hope to see you in 2021, in our next Computing Conference, with the same
amplitude, focus and determination.
Kohei Arai
vi Editor ’s Preface

Contents
Preventing Neural Network Weight Stealing
via Network Obfuscation
.................................... 1
Kálmán Szentannai, Jalal Al-Afandi, and András Horváth
Applications of Z-Numbers and Neural Networks in Engineering
.....12
Raheleh Jafari, Sina Razvarz, and Alexander Gegov
5G-FOG: Freezing of Gait Identiﬁcation in Multi-class Softmax
Neural Network Exploiting 5G Spectrum
........................ 26
Jan Sher Khan, Ahsen Tahir, Jawad Ahmad, Syed Aziz Shah,
Qammer H. Abbasi, Gordon Russell, and William Buchanan
Adaptive Blending Units: Trainable Activation Functions for Deep
Neural Networks
.......................................... 37
Leon RenéSütfeld, Flemming Brieger, Holger Finger, Sonja Füllhase,
and Gordon Pipa
Application of Neural Networks to Characterization
of Chemical Sensors
........................................ 51
Mahmoud Zaki Iskandarani
Application of Machine Learning in Deception Detection
............ 61
Owolafe Otasowie
A New Approach to Estimate the Discharge Coefﬁcient
in Sharp-Crested Rectangular Side Oriﬁces Using Gene
Expression Programming
.................................... 77
Hossein Bonakdari, Bahram Gharabaghi, Isa Ebtehaj, and Ali Shariﬁ
DiaTTroD: A Logical Agent Diagnostic Test for Tropical Diseases
....97
Sandra Mae W. Famador and Tardi Tjahjadi
A Weighted Combination Method of Multiple K-Nearest Neighbor
Classiﬁers for EEG-Based Cognitive Task Classiﬁcation
............116
Abduljalil Mohamed, Amer Mohamed, and Yasir Mustafa
vii

Detection and Localization of Breast Tumor in 2D Using
Microwave Imaging
........................................ 132
Abdelfettah Miraoui, LotﬁMerad Sidi, and Mohamed Meriah
Regression Analysis of Brain Biomechanics Under
Uniaxial Deformation
....................................... 142
O. Abuomar, F. Patterson, and R. K. Prabhu
Exudate-Based Classiﬁcation for Detection of Severity of Diabetic
Macula Edema
............................................ 150
Nandana Prabhu, Deepak Bhoir, Nita Shanbhag, and Uma Rao
Analysis and Detection of Brain Tumor Using U-Net-Based
Deep Learning
............................................ 161
Vibhu Garg, Madhur Bansal, A. Sanjana, and Mayank Dave
Implementation of Deep Neural Networks in Facial Emotion
Perception in Patients Suffering from Depressive Disorder: Promising
Tool in the Diagnostic Process and Treatment Evaluation
...........174
Krzysztof Michalik and Katarzyna Kucharska
Invisibility and Fidelity Vector Map Watermarking Based
on Linear Cellular Automata Transform
........................ 185
Saleh Al-Ardhi, Vijey Thayananthan, and Abdullah Basuhail
Implementing Variable Power Transmission Patterns
for Authentication Purposes
.................................. 198
Hosam Alamleh, Ali Abdullah S. Alqahtani, and Dalia Alamleh
SADDLE: Secure Aerial Data Delivery with Lightweight Encryption
...204
Anthony Demeri, William Diehl, and Ahmad Salman
Malware Analysis with Machine Learning for Evaluating the Integrity
of Mission Critical Devices
................................... 224
Robert Heras and Alexander Perez-Pons
Enhanced Security Using Elasticsearch and Machine Learning
.......244
Ovidiu Negoita and Mihai Carabas
Memory Incentive Provenance (MIP) to Secure the Wireless Sensor
Data Stream
.............................................. 255
Mohammad Amanul Islam
Tightly Close It, Robustly Secure It: Key-Based Lightweight Process
for Propping up Encryption Techniques
........................ 278
Muhammed Jassem Al-Muhammed, Ahmad Al-Daraiseh,
and Raed Abuzitar
viii Contents

Statistical Analysis to Optimize the Generation of Cryptographic Keys
from Physical Unclonable Functions
............................ 302
Bertrand Cambou, Mohammad Mohammadi, Christopher Philabaum,
and Duane Booher
Towards an Intelligent Intrusion Detection System:
A Proposed Framework
..................................... 322
Raghda Fawzey Hriez, Ali Hadi, and Jalal Omer Atoum
LockChain Technology as One Source of Truth for Cyber,
Information Security and Privacy
............................. 336
Yuri Bobbert and Nese Ozkanli
Introduction of a Hybrid Monitor for Cyber-Physical Systems
.......348
J. Ceasar Aguma, Bruce McMillin, and Amelia Regan
Software Implementation of a SRAM PUF-Based Password Manager
...361
Sareh Assiri, Bertrand Cambou, D. Duane Booher,
and Mohammad Mohammadinodoushan
Contactless Palm Vein Authentication Security Technique for Better
Adoption of e-Commerce in Developing Countries
................. 380
Sunday Alabi, Martin White, and Natalia Beloff
LightGBM Algorithm for Malware Detection
.................... 391
Mouhammd Al-kasassbeh, Mohammad A. Abbadi,
and Ahmed M. Al-Bustanji
Exploiting Linearity in White-Box AES with Differential
Computation Analysis
...................................... 404
Jakub Klemsa and Martin Novotný
Immune-Based Network Dynamic Risk Control Strategy Knowledge
Ontology Construction
...................................... 420
Meng Huang, Tao Li, Hui Zhao, Xiaojie Liu, and Zhan Gao
Windows 10 Hibernation File Forensics
......................... 431
Ahmad Ghafarian and Deniz Keskin
Behavior and Biometrics Based Masquerade Detection
Mobile Application
......................................... 446
Pranieth Chandrasekara, Hasini Abeywardana, Sammani Rajapaksha,
Sanjeevan Parameshwaran, and Kavinga Yapa Abeywardana
Spoofed/Unintentional Fingerprint Detection Using Behavioral
Biometric Features
......................................... 459
Ammar S. Salman and Odai S. Salman
Enabling Paratransit and TNC Services with Blockchain Based
Smart Contracts
........................................... 471
Amari N. Lewis and Amelia C. Regan
Contents ix

A Review of Cyber Security Issues in Hospitality Industry..........482
Neda Shabani and Arslan Munir
Extended Protocol Using Keyless Encryption Based on Memristors
....494
Yuxuan Zhu, Bertrand Cambou, David Hely, and Sareh Assiri
Recommendations for Effective Security Assurance
of Software-Dependent Systems
............................... 511
Jason Jaskolka
On Generating Cancelable Biometric Templates Using Visual
Secret Sharing
............................................ 532
Manisha and Nitin Kumar
An Integrated Safe and Secure Approach for Authentication and
Secret Key Establishment in Automotive Cyber-Physical Systems
.....545
Naresh Kumar Giri, Arslan Munir, and Joonho Kong
How Many Clusters? An Entropic Approach to Hierarchical
Cluster Analysis
........................................... 560
Sergei Koltcov, Vera Ignatenko, and Sergei Pashakhin
Analysis of Structural Liveness and Boundedness in Weighted
Free-Choice Net Based on Circuit Flow Values
................... 570
Yojiro Harie and Katsumi Wasaki
Classiﬁcation of a Pedestrian’s Behaviour Using Dual Deep
Neural Networks
.......................................... 581
James Spooner, Madeline Cheah, Vasile Palade, Stratis Kanarachos,
and Alireza Daneshkhah
Towards Porting Astrophysics Visual Analytics Services
in the European Open Science Cloud
........................... 598
Eva Sciacca, Fabio Vitello, Ugo Becciani, Cristobal Bordiu,
Filomena Bufano, Antonio Calanducci, Alessandro Costa, Mario Raciti,
and Simone Riggi
Computer Graphics-Based Analysis of Anterior Cruciate
Ligament in a Partially Replaced Knee
......................... 607
Ahmed Imran
An Assessment Algorithm for Evaluating Students Satisfaction
in e-Learning Environments: A Case Study
...................... 613
M. Caramihai, Irina Severin, and Ana Maria Bogatu
The Use of New Technologies in the Organization
of the Educational Process
................................... 622
Y. A. Daineko, N. T. Duzbayev, K. B. Kozhaly, M. T. Ipalakova,
Zh. M. Bekaulova, N. Zh. Nalgozhina, and R. N. Sharshova
x Contents

Design and Implementation of Cryptocurrency Price
Prediction System
.......................................... 628
Milena Karova, Ivaylo Penev, and Daniel Marinov
Strategic Behavior Discovery of Multi-agent Systems Based
on Deep Learning Technique
................................. 644
Boris Morose, Sabina Aledort, and Gal Zaidman
Development of Prediction Methods for Taxi Order Service
on the Basis of Intellectual Data Analysis
........................ 652
N. A. Andriyanov
Discourse Analysis on Learning Theories and AI
.................. 665
Rosemary Papa, Karen Moran Jackson, Ric Brown, and David Jackson
False Asymptotic Instability Behavior at Iterated Functions
with Lyapunov Stability in Nonlinear Time Series
................. 673
Charles Roberto Telles
The Inﬂuence of Methodological Tools on the Diagnosed Level
of Intellectual Competence in Older Adolescents
.................. 694
Sipovskaya Yana Ivanovna
The Automated Solar Activity Prediction System (ASAP) Update
Based on Optimization of a Machine Learning Approach
...........702
Ali K. Abed and Rami Qahwaji
Author Index
................................................ 719
Contents xi

Preventing Neural Network Weight
Stealing via Network Obfuscation
K´alm´an Szentannai, Jalal Al-Afandi, and Andr´as Horv´ath
(B)
Faculty of Information Technology and Bionics, Peter Pazmany Catholic University,
Pr´ater u. 50/A, Budapest 1083, Hungary
[email protected]
Abstract.Deep Neural Networks are robust to minor perturbations of
the learned network parameters and their minor modiﬁcations do not
change the overall network response signiﬁcantly. This allows space for
model stealing, where a malevolent attacker can steal an already trained
network, modify the weights and claim the new network his own intel-
lectual property. In certain cases this can prevent the free distribution
and application of networks in the embedded domain. In this paper, we
propose a method for creating an equivalent version of an already trained
fully connected deep neural network that can prevent network stealing,
namely, it produces the same responses and classiﬁcation accuracy, but
it is extremely sensitive to weight changes.
Keywords:Neural networks
·Networks stealing·Weight stealing·
Obfuscation
1 Introduction
Deep neural networks are employed in an emerging number of tasks, many of
which were not solvable before with traditional machine learning approaches. In
these structures, expert knowledge which is represented in annotated datasets is
transformed into learned network parameters known as network weights during
training.
Methods, approaches and network architectures are distributed openly in
this community, but most companies protect their data and trained networks
obtained from tremendous amount of working hours annotating datasets and
ﬁne-tuning training parameters.
Model stealing and detection of unauthorized use via stolen weights is a key
challenge of the ﬁeld as there are techniques (scaling, noising, ﬁne-tuning, distil-
lation) to modify the weights to hide the abuse, while preserving the functionality
and accuracy of the original network. Since networks are trained by stochastic
optimization methods and are initialized with random weights, training on a
dataset might result various diﬀerent networks with similar accuracy.
There are several existing methods to measure distances between network
weights after these modiﬁcations and independent trainings: [1–3] Obfuscation of
cﬀSpringer Nature Switzerland AG 2020
K. Arai et al. (Eds.): SAI 2020, AISC 1230, pp. 1–11, 2020.
https://doi.org/10.1007/978-3-030-52243-8
_1

2 K.Szentannaietal.
neural networks was introduced in [4], which showed the viability and importance
of these approaches. In this paper the authors present a method to obfuscate
the architecture, but not the learned network functionality. We would argue that
most ownership concerns are not raised because of network architectures, since
most industrial applications use previously published structures, but because of
network functionality and the learned weights of the network.
Other approaches try to embed additional, hidden information in the network
such as hidden functionalities or non-plausible, predeﬁned answers for previously
selected images (usually referred as watermarks) [5,6]. In case of a stolen network
one can claim ownership of the network by unraveling the hidden functionality,
which can not just be formed randomly in the structure. A good summary com-
paring diﬀerent watermarking methods and their possible evasions can be found
in [7].
Instead of creating evidence, based on which relation between the original
and the stolen, modiﬁed model could be proven, we have developed a method
which generates a completely sensitive and fragile network, which can be freely
shared, since even minor modiﬁcation of the network weights would drastically
alter the networks response.
In this paper, we present a method which can transform a previously trained
network into a fragile one, by extending the number of neurons in the selected
layers, without changing the response of the network. These transformations can
be applied in an iterative manner on any layer of the network, except the ﬁrst and
the last layers (since their size is determined by the problem representation). In
Sect.2we will ﬁrst introduce our method and the possible modiﬁcations on stolen
networks and in Sect.3we will describe our simulations and results. Finally in
Sect.4we will conclude our results and describe our planned future work.
2 Mathematical Model of Unrobust Networks
2.1 Fully Connected Layers
In this section we would like to present our method, how a robust network
can be transformed into a non-robust one. We have chosen fully connected net-
works because of their generality and compact mathematical representation.
Fully connected networks are generally applied covering the whole spectrum of
machine learning problems from regression through data generation to classi-
ﬁcation problems. The authors can not deny the fact, that in most practical
problems convolutional networks are used, but we would like to emphasize the
following properties of fully connected networks:(1)in those cases when there
is no topographic correlation in the data, fully connected networks are applied
(2)most problems also contain additional fully connected layers after the fea-
ture extraction of the convolutional or residual layers(3)every convolutional
network can be considered as a special case of fully connected ones, where all
weights outside the convolutional kernels are set to zero.
A fully connected deep neural network might consist of several hidden lay-
ers each containing certain number of neurons. Since all layers have the same

Preventing Neural NetworkWeight Stealing via Network Obfuscation 3
architecture, without the loss of generality, we will focus here only on three con-
secutive layers in the network (i−1,iandi+ 1). We will show how neurons
in layerican be changed, increasing the number of neurons in this layer and
making the network fragile, meanwhile keeping the functionality of the three
layers intact. We have to emphasize that this method can be applied on any
three layers, including the ﬁrst and last three layers of the network and also that
it can be applied repeatedly on each layer, still without changing the overall
functionality of the network.
The input of the layeri, the activations of the previous layer (i−1) can be
noted by the vectorx
i−1containingNelements. The weights of the network
are noted by the weight matrixW
iand the biasb iwhere W is a matrix of
N×Kelements, creating a mappingR
N
φ →R
K
andb iis a vector containingK
elements. The output of layeri, also the input of layeri+ 1 can be written as:
x
i=φ(W iN×K
xi−1+bi) (1)
whereφis the activation function of the neurons.
The activations of layeri+ 1 can be extended as using Eq.1:
x
i+1=φ(φ(xW i−1N×K
+bi−1)WiK×L
+bi) (2)
Creating a mappingR
N
φ →R
L
.
One way of identifying a fully connected neural network is to represent it as a
sequence of synaptic weights. Our assumption was that in case of model stealing
certain application of additive noise on the weights would prevent others to reveal
the attacker and conceal thievery. Since fully connected networks are known to be
robust against such modiﬁcations, the attacker could use the modiﬁed network
with approximately the same classiﬁcation accuracy. Thus, our goal was to ﬁnd
a transformation that preserves the loss and accuracy rates of the network, but
introduces a signiﬁcant decrease in terms of the robustness against parameter
tuning. In case of a three-layered structure one has to preserve the mapping
between the ﬁrst and third layers (Eq.2) to keep the functionality of this three
consecutive layers, but the mapping in Eq.1(the mapping between the ﬁrst and
second, or second and third layers), can be changed freely.
Also, our model must rely on an identiﬁcation mechanism based on a repre-
sentation of the synaptic weights. Therefore, the owner of a network should be
able to verify the ownership based on the representation of the neural network,
examining the average distance between the weights [7].
2.2 Decomposing Neurons
We would like to ﬁnd suchW
α
i−1
N×M
andW
α
i
M×L
(M∈N,M >K) matrices, for
which:
φ(φ(xW
i−1N×K
+bi−1)WiK×L
+bi)
=φ(φ(xW
α
i−1
N×M
+b
α
i−1
)W
α
i
M×L
+bi)
(3)

4 K.Szentannaietal.
Considering the linear case whenφ(x)=xwe obtain the following form:
xW
i−1N×K
WiK×L
+bi−1WiK×L
+bi
=xW
α
i−1
N×M
W
α
i
M×L
+b
α
i−1
W
α
i
M×L
+bi
(4)
The equation above holds only for the special case ofφ(x)=x, however in
most cases nonlinear activation functions are used. We have selected the rectiﬁed
linear unit (ReLU) for our investigation (φ(x)=max(0,x)). This non-linearity
consist of two linear parts, which means that a variable could be in a linear
domain of Eq.3resulting selected lines of4(ifx≥0), or the equation system
is independent from the variable if the activation function results a constant
zero (ifx≤0). This way ReLU gives a selection of given variables (lines) of4.
However, applying the ReLU activation function has certain constraints.
Assume, that a neuron with the ReLU activation function should be replaced
by two other neurons. This can be achieved by using anα∈(0,1) multiplier:
φ(
n
φ
i=1
W
l
ji
xi+b
l
j
)=N
l
j
(5)
N
l
j
=αN
l
j
+(1−α)N
l
j
(6)
whereαN
l
j
and (1−α)N
l
j
correspond to the activation of the two neurons.
For each of these, the activation would only be positive if the original neuron
had a positive activation, otherwise it would be zero, this means that all the
decomposed neuron must have the same bias.
After decomposing a neuron, it is needed to choose the appropriate weights
on the subsequent layer. A trivial solution is to keep the original synaptic weights
represented by the
W
l+1
j
column vector. This would lead to the same activation
since
N
l
j
W
l+1
j
=αN
l
j
W
l+1
j
+(1−α)N
l
j
W
l+1
j
(7)
A fragile network can be created by choosing the same synaptic weights for
the selected two neurons, but it would be easy to spot by the attacker, thus
another solution is needed. In order to ﬁnd a nontrivial solution we constructed
a linear equation system that can be described by equation systemAp=c,
whereAcontains the original, already decomposed synaptic weights of the ﬁrst
layer, meanwhile,prepresents the unknown synaptic weights of the subsequent
layer. Vectorccontains the corresponding weights from the original network
multiplied together: each element represents the amount of activation related
to one input. Finally the non-trivial solution can be obtained by solving the
following non-homogeneous linear equation system for each output neuron where
indexjdenotes the output neuron.
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎣
w
1
φ
11w
1
φ
21... w
1
φ
m1
w
1
φ
12w
1
φ
22... w
1
φ
m2
.
.
.
.
.
.
.
.
.
.
.
.
w
1
φ
1nw
1
φ
2n... w
1
φ
mn
b
1
φ
1b
1
φ
2... b
1
φ
m
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎦
×
⎡
⎢
⎢
⎢
⎢
⎣
w
2
φ
j1
w
2
φ
j2
.
.
.
w
2
φ
jm
⎤
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎣

k
i=1
w
2
ji
w
1
i1

k
i=1
w
2
ji
w
1
i2
.
.
.

k
i=1
w
2
ji
w
1
ik

k
i=1
w
2
ji
b
1
i
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎦
(8)

Preventing Neural NetworkWeight Stealing via Network Obfuscation 5
It is important to note, that all the predeﬁned weights on layerl+ 1 might
change. In summary, this step can be considered as the replacement of a layer,
changing all synaptic weights connecting from and to this layer, but keeping the
biases of the neurons and the functionality of the network intact.
The only constraint of this method is related to the number of neurons regard-
ing the two consecutive layers. It is known, that for matrixAwith the size of
M×N, equationAp=chas a solution if and only ifrank(A)=rank[A|c] where
[A|c] is the extended matrix. The decomposition of a neuron described in Eq.7
results in linearly dependent weight vectors on layerl, therefore when solving the
equation system the rank of the matrixAis less than or equal tomin(N+1,K).
If the rank is equal toN+ 1 (meaning thatK≥N+ 1) then vectorcwith the
dimension ofN+1 would not introduce a new dimension to the subspace deﬁned
by matrixA. However ifrank(A)=K(meaning thatK≤N+ 1) then vector
ccould extend the subspace deﬁned byA. Therefore, the general condition for
solving the equation system is:K≥N+1.
This shows that one could increase the number of the neurons in a layer,
and divide the weights of the existing neuron in that layer. We have used this
derivation and aim to ﬁnd a solution of Eq.7where the order of magnitudes are
signiﬁcantly diﬀerent (in the range of 10
6
) for both the network parameters and
for the eigenvalues of the mappingR
N
φ →R
L
.
2.3 Introducing Deceptive Neurons
The method described in the previous section results a fragile neural network, but
unfortunately it is not enough to protect the network weights, since an attacker
could identify the decomposed neurons based on their biases or could ﬁt a new
neural network on the functionality implemented by the layer. To prevent this
we will introduce deceptive neurons in layers. The purpose of these neurons is to
have non-zero activation in sum if and only if noise was added to their weights
apart from this all these neurons have to cancel each others eﬀect out in the
network, but not necessarily in a single layer.
The simplest method is to deﬁne a neuron with an arbitrary weight and a bias
of an existing neuron resulting a large activation and making a copy of it with
the diﬀerence of multiplying the output weights by−1. As a result, these neurons
do not contribute to the functionality of the network. However, adding noise to
the weights of these neurons would have unexpected consequences depending on
the characteristics of the noise, eventually leading to a decrease of classiﬁcation
accuracy.
One important aspect of this method is to hide the generated neurons and
obfuscate the network to prevent the attacker to easily ﬁlter our deceptive neu-
rons in the architecture. Choosing the same weights again on both layers would
be an obvious sign to an attacker, therefore this method should be combined
with decomposition described in Sect.2.2.
Since decomposition allows the generation of arbitrarily small weights one
can select a suitably small magnitude, which allows the generation ofRreal
(non deceptive) neurons in the system, and half of their weights (αparameters)

6 K.Szentannaietal.
can be set arbitrarily, meanwhile the other half of the weights will be determined
by Eq.8. For each real neuron one can generate a number (F) of fake neurons
forming groups ofRnumber of real andFnumber of fake neurons. These groups
can be easily identiﬁed in the network since all of them will have the same bias,
but the identiﬁcation of fake and real neurons in a group is non-polynomial.
The eﬃciency of this method should be measured in the computational com-
plexity of successfully ﬁnding two or more corresponding fake neurons having
a total activation of zero in a group. Assuming that only one pair of fake neu-
rons was added to the network, it requires
α
L
i=0
∈
Ri+Fi
2
≥
steps to successfully
identify the fake neurons, whereR
i+Fidenotes the number of neurons in the
corresponding hidden layer, andLis the number of hidden layers. This can be
further increased by decomposing the fake neurons using Eq.8:inthatcasethe
required number of steps is
α
L
i=0
∈
Ri+Fi
d+2
≥
,dbeing the number of extra decom-
posed neurons. This can be maximized ifd+2 =R
i+Fi/2, whereidenotes
the layer, where the fake neurons are located. However, this is true only if the
attacker has information about the number of deceptive neurons. Without any
prior knowledge, the attacker has to guess the number of deceptive neurons
as well (0,1,2...R
i+Fi−1) which leads to exponentially increasing computa-
tional time.
3 Experiments
3.1 Simulation of a Simple Network
As a case study we have created a simple fully connected neural network with
three layers, each containing two neurons to present the validity of our approach.
The functionality of the network can be considered as a mappingf:R
2
φ →R
2
.
w
1=
≤
6−1
−17

,b 1=

1−5

w 2=
≤
53
9−1

,b 2=

71

We added two neurons to the hidden layer with decomposition, which does
not modify the input and output space and no deceptive neurons were used in
this experiment. After applying the methods described in Sect.2.1, we obtained
a solution of:
w
1=
≤
0.0525−0.4213 6.0058−0.5744
−0.0087 2.9688−0.9991 4.0263

b
1=

0.0087−2.1066 1.0009−2.8722

w
2=
⎡
⎢
⎢
⎣
4.1924e+03−5.4065e+03
−2.3914 7 .3381
−3.2266 5 .7622
6.9634 −7.0666
⎤
⎥
⎥
⎦
b
2=

71

Preventing Neural NetworkWeight Stealing via Network Obfuscation 7
Fig. 1.This ﬁgure depicts the response of a simple two-layered fully connected network
for a selected input (red dot) and the response of its variants with %1 noise (yellow
dots) added proportionally to the weights. The blue dots represent the responses of the
transformed MimosaNets under the same level of noise on their weights, meanwhile the
response of the transformed network (without noise) remained exactly the same.
In the following experiment we have chosen an arbitrary input vector: [7,9].
We have measured the response of the network for this input, each time intro-
ducing 1% noise to the weights of the network. Figure1shows the response of the
original network and the modiﬁed network after adding 1% noise. The variances
of the original network for the ﬁrst output dimension is 6.083 and 8.399 for the
second, meanwhile the variances are 476.221 and 767.877 for the decomposed
networks respectively. This example demonstrates how decomposition of a layer
can increase the networks dependence on its weights.
3.2 Simulations on the MNIST Dataset
We have created a ﬁve layered fully connected network containing 32 neurons in
each hidden-layer (and 728 and 10 neurons in the input and output layers) and
trained it on the MNIST [8] dataset, using batches of 32 and Adam Optimizer
[9] for 7500 iterations. The network has reached an accuracy of 98.4% on the
independent test set.
We have created diﬀerent modiﬁcations of the network by adding 9,18,36,72
extra neurons. These neurons were divided equally between the three hidden-
layers and 2/3 of them were deceptive neurons (since they were always created
in pairs) and 1/3 of them were created by decomposition. This means that in
case of 36 additional neurons 2×4 deceptive neurons were added to each layer
and four new neurons per layer were created by decomposition.

8 K.Szentannaietal.
In our hypothetical situations these networks (along with the original) could
be stolen by a malevolent attacker, who would try to conceal his thievery by
using the following three methods: adding additive noise proportionally to the
network weights, continuing network training on arbitrary data and network
knowledge distillation. All reported datapoints are an average of 25 independent
measurements.
Dependence on Additive Noise.We have investigated network performance
using additive noise to the network weights. The decrease of accuracy which
depends on the ratio of the additive noise can be seen in Fig.2.
At ﬁrst we have tested a fully connected neural network trained on the
MNIST dataset without making modiﬁcations to it. The decrease of accuracy
was not more than 0.2% even with a relatively high 5% noise. This shows the
robustness of a fully connected network.
After applying the methods described in Sect.2network accuracy retro-
gressed to 10% even in case of noise which was less than 1% of the network
weights, as Fig.2depicts. This alone would reason the applicability of our
method, but we have investigated low level noises further, which can be seen
on Fig.3. As it can be seen from the ﬁgure, accuracy starts to drop when the
ratio of additive noise reaches the level of 10
−7
, which means the attacker can
not signiﬁcantly modify the weights. This eﬀect could be increased by adding
more and more neurons to the network.
Fig. 2.This ﬁgure depicts accuracy changes on the MNIST dataset under various level
of additive noise applied on the weights. The original network (purple) is not dependent
on these weight changes, meanwhile accuracies retrogress in the transformed networks,
even with the lowest level of noise.
Dependence on Further Training Steps. Additive noise randomly modi-
ﬁes the weights, but it is important to examine how accuracy changes in case
of structured changes exploiting the gradients of the network. Figure4depicts
accuracy changes and average in weights distances by applying further training

Preventing Neural NetworkWeight Stealing via Network Obfuscation 9
steps in the network. Further training was examined using diﬀerent step sizes
and optimizers (SGD,AdaGrad and ADAM) training the network with original
MNIST and randomly selected labels and the results were qualitatively the same
in all cases.
Dependence on Network Distillation.We have tried to distill the knowledge
in the network and train a new neural network to approximate the functionality
of previously selected layers, by applying the method described in [10].
We have generated one million random input samples with their outputs for
the modiﬁed networks and have used this dataset to approximate the function-
ality of the network.
Fig. 3.A logarithmic plot depicting the same accuracy dependence as on Fig.2, focus-
ing on low noise levels. As it can be seen from the plot, accuracy values do not change
signiﬁcantly under 10
−7
percent of noise, which means the most important values of
the weights would remain intact to proof connection between the original and modiﬁed
networks.
We have created three-layered neural networks containing 32, 48, 64, 128
neurons in the hidden layer (The number of neurons in the ﬁrst and last layer
were determined by the original network) and tried to approximate the function-
ality of the hidden layers of the original structure. Since deceptive neurons have
activations in the same order of magnitude as the original responses, these values
disturb the manifold of the embedded representations learned by the network
and it is more diﬃcult to be approximated by a neural network. Table1contains
the maximum accuracies which could be reached with knowledge distillation,
depending on the number of deceptive neurons and the neurons in the architec-
ture used for distillation. This demonstrates, that our method is also resilient
towards knowledge distillation.

10 K. Szentannai et al.
Fig. 4.The ﬁgure plots accuracy dependence of the networks in case of further training
(applying further optimization steps). As it can be seen from the plot weights had to
be kept in 10
−7
average distance to keep the same level of accuracy.
Table 1.The table displays the maximum accuracies reached with knowledge distil-
lation. The diﬀerent rows display the number of extra neurons which were added to
the investigated layer, and the diﬀerent columns show the number of neurons in the
hidden layer of the fully connected architecture, which was used for distillation.
#Deceptive N.#N.=32 #N.=48 #N. = 64#N. = 128
9 0.64 0.65 0.69 0.71
18 0.12 0.14 0.15 0.17
36 0.10 0.11 0.10 0.13
72 0.11 0.09 0.10 0.10
4 Conclusion
In this paper, we have shown a transformation method which can signiﬁcantly
increase a network’s dependence on its weights, keeping the original functionality
intact. We have also presented how deceptive neurons can be added to a network,
without disturbing its original response. Using these transformations iteratively
one can create and openly share a trained network, where it is computationally
extensive to reverse engineer the original network architecture and embeddings
in the hidden layers. The drawback of the method is the additional computa-
tional need for the extra neurons, but this is not signiﬁcant, since computational
increase is polynomial.

Preventing Neural NetworkWeight Stealing via Network Obfuscation 11
We have tested our method on simple toy problems and on the MNIST
dataset using fully-connected neural networks and demonstrated that our app-
roach results non-robust networks for the following perturbations: additive noise,
application of further training steps and knowledge distillation.
Acknowledgments. This research has been partially supported by the Hungarian
Government by the following grant: 2018-1.2.1-NKP-00008: Exploring the Mathemat-
ical Foundations of Artiﬁcial Intelligence also the funds of grant EFOP-3.6.2-16-2017-
00013 are gratefully acknowledged.
References
1. Koch, E., Zhao, J.: Towards robust and hidden image copyright labeling. In: IEEE
Workshop on Nonlinear Signal and Image Processing, vol. 1174, pp. 185–206,
Greece, Neos Marmaras (1995)
2. Wolfgang, R.B., Delp, E.J.: A watermark for digital images. In: Proceedings of the
International Conference on Image Processing, vol. 3, pp. 219–222. IEEE (1996)
3. Zarrabi, H., Hajabdollahi, M., Soroushmehr, S., Karimi, N., Samavi, S., Najarian,
K.: Reversible image watermarking for health informatics systems using distortion
compensation in wavelet domain (2018 )arXiv preprintarXiv:1802.07786
4. Xu, H., Su, Y., Zhao, Z., Zhou, Y., Lyu, M.R., King, I.: Deepobfuscation: securing
the structure of convolutional neural networks via knowledge distillation (2018)
arXiv preprintarXiv:1806.10313
5. Namba, R., Sakuma, J.: Robust watermarking of neural network with exponential
weighting (2019)arXiv preprintarXiv:1901.06151
6. Gomez, L., Ibarrondo, A., M´arquez, J., Duverger, P.: Intellectual property protec-
tion for distributed neural networks (2018)
7. Hitaj, D., Mancini, L.V.: Have you stolen my model? evasion attacks against deep
neural network watermarking techniques (2018)arXiv preprintarXiv:1809.00615
8. LeCun, Y., Cortes, C., Burges, C.: MNIST handwritten digit database. AT&T
Labs2(2010).http://yann.lecun.com/exdb/mnist
9. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization (2014)arXiv
preprintarXiv:1412.6980
10. Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network
(2015)arXiv preprintarXiv:1503.02531

Applications of Z-Numbers and Neural
Networks in Engineering
Raheleh Jafari
1(B)
,SinaRazvarz
2
, and Alexander Gegov
3
1
School of design, University of Leeds, Leeds LS2 9JT, UK
[email protected]
2
Departamento de Control Autom´atico, CINVESTAV-IPN (National Polytechnic
Institute), Mexico City, Mexico
[email protected]
3
School of Computing, University of Portsmouth, Buckingham Building, PO1 3HE
Portsmouth, UK
[email protected]
Abstract.In the real world, much of the information on which deci-
sions are based is vague, imprecise and incomplete. Artiﬁcial intelligence
techniques can deal with extensive uncertainties. Currently, various types
of artiﬁcial intelligence technologies, like fuzzy logic and artiﬁcial neural
network are broadly utilized in the engineering ﬁeld. In this paper, the
combined Z-number and neural network techniques are studied. Further-
more, the applications of Z-numbers and neural networks in engineering
are introduced.
Keywords:Artiﬁcial intelligence
·Fuzzy logic·Z-number·Neural
network
1 Introduction
Intelligent systems are composed of fuzzy systems and neural networks. They
have particular properties such as the capability of learning, modeling and resolv-
ing optimizing problems, suitable for speciﬁc kind of applications. The intelligent
system can be named hybrid system in case that it combines a minimum of two
intelligent systems. For example, the mixture of the fuzzy system and neural
network causes the hybrid system to be called a neuron-fuzzy system.
Neural networks are made of interrelated groups of artiﬁcial neurons that
have information which is obtainable by computations linked to them. Mostly,
neural networks can adapt themselves to structural alterations while the training
phase. Neural networks have been utilized in modeling complicated connections
among inputs and outputs or acquiring patterns for the data [1–12].
Fuzzy logic systems are broadly utilized to model the systems characterizing
vague and unreliable information [13–29]. During the years, investigators have
proposed extensions to the theory of fuzzy logic. Remarkable extension includes
Z-numbers [30]. The Z-number is deﬁned as an ordered pair of fuzzy numbers
cﬀSpringer Nature Switzerland AG 2020
K. Arai et al. (Eds.): SAI 2020, AISC 1230, pp. 12–25, 2020.
https://doi.org/10.1007/978-3-030-52243-8
_2

Applications of Z-Numbers and Neural Networks in Engineering 13
(C, D), such thatCis a value of some variables andDis the reliability which
is a value of probability rate ofC. Z-numbers are widely applied in various
implementations in diﬀerent areas [31–36].
In this paper, the basic principles and explanations of Z-numbers and neu-
ral networks are given. The applications of Z-numbers and neural networks in
engineering are introduced. Also, the combined Z-number and neural network
techniques are studied. The rest of the paper is organized as follows. The the-
oretical background of Z-numbers and artiﬁcial neural networks are detailed in
Sect.2. Comparison analysis of neural networks and Z-number systems is pre-
sented in Sect.3. The combined Z-number and neural network techniques are
given in Sect.4. The conclusion of this work is summarized in Sect.5.
2 Theoretical Background
In this section, we provide a brief theoretical insight of Z-numbers and artiﬁcial
neural networks.
2.1 Z-Numbers
Mathematical Preliminaries.Here some necessary deﬁnitions of Z-number
theory are given.
Deﬁnition 1.Ifqis: 1) normal, there existsω
0ωυwhereq(ω 0) = 1, 2) convex,
q(υω+(1−υ)ω)≥min{q(ω),q(τ)},∀ω, τωυ,∀υ∈[0,1],3) upper semi-
continuous onυ,q(ω)≤q(ω
0)+⎧,∀ω∈N(ω 0),∀ω 0ωυ,∀⎧>0,N(ω 0)isa
neighborhood, 4)q
+
={ωωυ,q(ω)>0}is compact, soqis a fuzzy variable,
q∈E:υζ[0,1].
The fuzzy variableqis deﬁned as below
q=
ω
q
,q
υ
(1)
such thatqis the lower-bound variable andqis the upper-bound variable.
Deﬁnition 2.TheZ-number is composed of two elementsZ=[q(ω),p].q(ω)
is considered as the restriction on the real-valued uncertain variableωandpis
considered as a measure of the reliability ofq.TheZ-number is deﬁned asZ
+
-
number, whenq(ω) is a fuzzy number andpis the probability distribution ofω.
Ifq(ω), andp, are fuzzy numbers, then theZ-number is deﬁned asZ
−
-number.
TheZ
+
-number has more information in comparison with theZ
−
-number.
In this work, we use the deﬁnition ofZ
+
-number, i.e.,Z=[q, p],qis a fuzzy
number andpis a probability distribution.
The triangular membership function is deﬁned as
μ
q=G(a, b, c)=
τ
ω−a
b−a
a≤ω≤b
c−ω
c−b
b≤ω≤c
otherwise μ
q= 0 (2)

14 R. Jafari et al.
and the trapezoidal membership function is deﬁned as
μ
q=G(a, b, c, d)=
⎧
⎨
⎩
ω−a
b−a
a≤ω≤b
d−ω
d−c
c≤ω≤d
1b≤ω≤c
otherwise μ q= 0 (3)
The probability measure ofqis deﬁned as
P(q)=
ϑ
υ
μq(ω)p(ω)dω (4)
such thatpis the probability density ofω. For discreteZ-numbers we have
P(q)=
n

j=1
μq(ωj)p(ω j) (5)
Deﬁnition 3.Theα-level of theZ-numberZ=(q, p) is stated as below
[Z]
α
=([q]
α
,[p]
α
) (6)
such that 0<α≤1. [p]
α
is calculated by the Nguyen’s theorem
[p]
α
=p([q]
α
)=p([q
α
,q
α
]) =

P
α
,P
α

(7)
such thatp([q]
α
)={p(ω)|ω∈[q]
α
}. Hence, [Z]
α
is deﬁned as
[Z]
α
=

Z
α
,Z
α

=

ω
q
α
,P
α
υ
,

q
α
,P
α

(8)
such thatP
α
=q
α
p(ω
α
j
),
P
α
=q
α
p(ω
α
j
), [ωj]
α
=(ω
α
j
,
ω
α
j
).
LetZ
1=(q1,p1)andZ 2=(q2,p2), we have
Z
12=Z1∗Z2=(q1∗q2,p1∗p2) (9)
where∗∈{⊕,,}.⊕,and, indicate sum, subtract and multiply, respec-
tively.
The operations utilized for the fuzzy numbers [q
1]
α
=[q
α
11
,q
α
12
] and [q 2]
α
=
[q
α
21
,q
α
22
] are deﬁned as [37],
[q
1⊕q2]
α
=[q1]
α
+[q2]
α
=[q
α
11
+q
α
21
,q
α
12
+q
α
22
]
[q
1q2]
α
=[q1]
α
−[q2]
α
=[q
α
11
−q
α
22
,q
α
12
−q
α
21
]
[q
1q2]
α
=

min{q
α
11
q
α
21
,q
α
11
q
α
22
,q
α
12
q
α
21
,q
α
12
q
α
22
}
max{q
α
11
q
α
21
,q
α
11
q
α
22
,q
α
12
q
α
21
,q
α
12
q
α
22
}
(10)
For the discrete probability distributions, the following relation is deﬁned for all
p
1∗p2operations
p
1∗p2=

ι
p1(ω1,j)p2(ω
2,(n−j) )=p 12(ω) (11)

Applications of Z-Numbers and Neural Networks in Engineering 15
Fig. 1.Membership functions applied for (a) cereal yield, cereal production, economic
growth, (b) threat rate, and (c) reliability
Background and Related Work. The implementations of Z-numbers based
techniques are bounded because of the shortage of eﬀective approaches for cal-
culation with Z-numbers.
In [38], the capabilities of the Z-numbers in the improvement of the quality
of risk assessment are studied. Prediction equal to (High, Very Sure) is institu-
tionalized in the form of Z-evaluation “yisZ(c, p)”, such thatyis considered
as a random variable of threat probability,candpare taken to be fuzzy sets,
demonstrating soft constraints on a threat probability and a partial reliability,
respectively. The likelihood of risk is illustrated by Z-number as: Probability
=Z
1(High, Very Sure), such thatcis indicated through linguistic terms High,
Medium, Low, also,pis indicated through terms Very Sure, Sure, etc. Likewise,
consequence rate is explained as: Consequence measure =Z
2(Low, Sure). Threat
rates (Z
12) is computed as the product of the probability (Z 1) and consequence
measure (Z
2).
In [39], Z-number-based fuzzy system is suggested to determine the food
security risk level. The proposed system is relying on fuzzy If-Then rules, which
applies the basic parameters such as cereal production, cereal yield, and economic
growth to specify the threat rate of food security. The membership functions
applied to explain input, as well as output variables, are demonstrated in Fig.1.
In [40], the application of the Z-number theory to selection of optimal
alloy is illustrated. Three alloys named Ti12Mo2Sn alloy, Ti12Mo4Sn alloy, and
Ti12Mo6Sn alloy are examined and an optimal titanium alloy is selected using
the proposed approach. The optimality of the alloys is studied based on three
criteria: strength level, plastic deformation degree, and tensile strength.

16 R. Jafari et al.
Fig. 2.The structure of a biological neuron
2.2 Neural Networks
Neural networks are constructed from neurons and synapses. They alter their
rates in reply from nearby neurons as well as synapses. Neural networks operate
similar to computer as they map inputs to outputs. Neurons, as well as synapses,
are silicon members, which mimic their treatment. A neuron gathers the total
incoming signals from other neurons, afterward simulate its reply represented
by a number. Signals move among the synapses, which contain numerical rates.
Neural networks learn once they vary the value of their synapsis. The structure
of a biological neuron or nerve cell is shown in Fig.2. The processing steps inside
each neuron is demonstrated in Fig.3.
Background and Related Work. In [41], artiﬁcial neural network technique
is utilized for modeling the void fraction in two-phase ﬂow inside helical vertical
coils with water as work ﬂuid. In [42] artiﬁcial neural network and multi-objective
genetic algorithm are applied for optimizing the subcooled ﬂow boiling in a
vertical pipe. Pressure, the mass ﬂux of the water, inlet subcooled temperature,
as well as heat ﬂux are considered as inlet parameters. The artiﬁcial neural
network utilizes inlet parameters for predicting the objective functions, which
are the maximum wall surface temperature as well as averaged vapor volume
fraction at the outlet. The optimization procedure of design parameters is shown
in Fig.4.
In [43], artiﬁcial neural network technique is applied for predicting heat trans-
fer in supercritical water. The artiﬁcial neural network is trained on the basis
of 5280 data points gathered from experimental results. Mass ﬂux, heat ﬂux,
pressure, tube diameter, as well as bulk speciﬁc enthalpy are taken to be the
inputs of the proposed artiﬁcial neural network. The tube wall temperature is
taken to be the output, see Fig.5.

Applications of Z-Numbers and Neural Networks in Engineering 17
Fig. 3.Processing steps inside each neuron
3 Comparison Analysis of Neural Networks
and Z-Number Systems
Neural networks and Z-number systems can be considered as a part of the soft
computing ﬁeld. The comparison of Neural networks and Z-number systems is
represented in Table1. Neural networks have the following advantageous:
Table 1.The comparison of neural networks and Z-number systems.
Z-number systemsNeural networks
Knowledge presentation Very good Very bad
Uncertainty tolerance Very good Very good
Inaccuracy tolerance Very good Very good
Compatibility Bad Very good
Learning capability Very bad Very good
Interpretation capability Very good Very bad
Knowledge detection and data miningBad Very good
Maintainability Good Very good
iAdaptive Learning: capability in learning tasks on the basis of the data sup-
plied to train or initial experience.
iiSelf-organization: neural networks are able to create their organization while
time learning.
iiiReal-time execution: the calculations of neural networks may be executed in
parallel, also speciﬁc hardware devices are constructed, which can capture
the beneﬁt of this feature.
Neural networks have the following drawbacks:

18 R. Jafari et al.
iThe utilization of neural networks is in direct connection with the availability
of the training data.
iiThe acquired solution from the learning procedure may not be often
explained.
iiiAlmost all the neural network systems contain black boxes such that the
ultimate state may not be explained.
Fuzzy logic has the following advantageous:
iSimple to learn and apply.
iiA user-friendly procedure to produce.
iiiGeneration of more eﬀective performance.
Fuzzy logic has the following drawbacks:
iConstructing an uncertain system is complex.
iiIt is not easy to deﬁne proper membership values for uncertain systems.
Fig. 4.The optimization procedure of input parameters

Applications of Z-Numbers and Neural Networks in Engineering 19
4 Combined Z-Number and Neural Network Techniques
4.1 Why Apply Z-Numbers in Neural Networks
Each neuron in the artiﬁcial neural network is linked with another neuron via a
connection link in such a manner that the connecting link is related to a weight
with the information regarding the input signal. Therefore, the weights contain
beneﬁcial information regarding input to resolve the problems. Some reasons for
applying Z-numbers in neural networks are as follows:
iIn a case that crisp values cannot be implemented, uncertain values such as
Z-numbers are utilized.
iiSince the training, as well as learning, assist neural network to have a high
performance in unanticipated status, therefore in such status, uncertain values
like Z-numbers are more suitable than crisp values.
iiiIn neural networks, Z-numbers are more applicable than fuzzy numbers.
Z-numbers are more precise when compared with fuzzy numbers. Also, Z-
numbers have less diﬃculty in computation in comparison with nonlinear
system modeling approaches.
Fig. 5.Proposed artiﬁcial neural network for predicting heat transfer in supercritical
water
4.2 Complexity in Applying Z-Numbers in Neural Networks
There exist some troubles when utilizing Z-numbers in neural networks. The
complexity is associated with membership rules, the requirement to construct
an uncertain system since it is often diﬃcult to derive it by supplied set of
complicated data.
Neural networks can be used to train Z-numbers. The advantageous of using
neural networks for training Z-numbers are as follows:
iNovel patterns of data may be learned simply using neural networks therefore,
it may be utilized for preprocessing data in uncertain systems.
iiNeural networks due to their abilities in learning new relation with new input
data may be utilized for reﬁning fuzzy rules to generate the fuzzy adaptive
system.

20 R. Jafari et al.
4.3 Examples of Combined Z-Number and Neural Network
Techniques
Example 1.The following system is designed such that inputs and outputs are
in the form of Z-numbers [44],
ζ(t)=ϑcos(ϕΔkt)
v(t+1)=
Δk
2
[ζ(t)−ψv
3
(t)]−v(t−1) +ρv(t)
(1 +ωΔk)
(12)
Fig. 6.Approximated error of multi-layer neural network
such thatρ=ωΔk−θΔk
2
+2.Δk,ω,θ,ψ,ϑare Z-number parameters.ϕis
taken to be a random variable uniformly distributed in the interval [0.1,2.9]
with meanE{ϕ}=1.5, as well as the initial conditions beingv(0) =v(1) = 1.
The following are assumed,
Δk= [(0.03,0.05,0.06),p(0.6,0.8,0.86)]
ω= [(0.1,0.3,0.5),p(0.6,0.7,0.87)]
θ=[(−4.2,−4,−3.8),p(0.6,0.8,86)]
ψ= [(0.8
,1,1.2),p(0.7,0.8,0.85)]
ϑ= [(0.2,0.5,0.7),p(0.7,0.8,0.85)]
(13)
In order to model the uncertain nonlinear system (12), a multi-layer neural
network is used such that obtains the Z-number coeﬃcients of (12). The error
plot is demonstrated in Fig.6.
Example 2.A liquid tank system is demonstrated in Fig.7, which is modeled as
below
d
dt
v(t)=−
1
SO
v(t)+
d
S
(14)
whered=t+ 1 is inﬂow disturbances of the tank that generates vibration in
liquid levelv,O= 1 is the ﬂow obstruction which can be curbed utilizing the

Applications of Z-Numbers and Neural Networks in Engineering 21
valve, alsoS= 1 is the cross-section of the tank. Two types of neural networks,
static Bernstein neural network (SBNN) and dynamic Bernstein neural network
(DBNN) [45], are used to estimate the Z-number solutions of (14). The error
plots of SBNN and DBNN are demonstrated in Fig.8.
Fig. 7.Liquid tank system
Fig. 8.Approximated errors of SBNN and DBNN
Example 3.The heat source by insulating materials is demonstrated in Fig.9,
which is modeled as below
M
1
N1
⊕
M
2
N2
=
M
3
N3
⊕
M
4
N4
⊕J (15)
A heat source is placed in the center of insulating materials. The widths of the
insulating materials are in the form of Z-numbers. The coeﬃcients of conductiv-
ity materials areN
1=h, N2=h
√
h, N3=h
2
,N4=
√
h, such thathis elapsed
time.Jis thermal resistance. Neural network technique is used to approximate
Z-number solutions of (15)[46].

22 R. Jafari et al.
Fig. 9.The heat source
5 Conclusion
The notion of Z-numbers is rather naturally obtained while gathering vague
information in a linguistic appearance. In this paper, the combined Z-number
and neural network techniques are studied. Furthermore, the applications of Z-
numbers and neural networks in engineering are introduced. As some researchers
have eﬀectively used Z-numbers, in-depth discussions are given for stimulating
future studies.
References
1. Dote, Y., Hoft, R.G.: Intelligent ControlPower Electronics Systems. Oxford Uni-
vsity Press, Oxford (1998)
2. Mohanty, S.: Estimation of vapour liquid equilibria for the system carbon dioxided-
iﬂuoromethane using artiﬁcial neural networks. Int. J. Refrig.29, 243249 (2006)
3. Razvarz, S., Jafari, R., Yu, W., Khalili, A.: PSO and NN modeling for photocat-
alytic removal of pollution in wastewater. In: 14th International Conference on
Electrical Engineering, Computing Science and Automatic Control (CCE) Electri-
cal Engineering, pp. 1–6 (2017)
4. Jafari, R., Yu, W.: Artiﬁcial neural network approach for solving strongly degen-
erate parabolic and burgers-ﬁsher equations. In: 12th International Conference on
Electrical Engineering, Computing Science and Automatic Control (2015).https://
doi.org/10.1109/ICEEE.2015.7357914
5. Jafari, R., Razvarz, S., Gegov, A.: A new computational method for solving fully
fuzzy nonlinear systems. In: Computational Collective Intelligence, ICCCI 2018.
Lecture Notes in Computer Science, vol. 11055, pp. 503–512. Springer, Cham
(2018)

Applications of Z-Numbers and Neural Networks in Engineering 23
6. Razvarz, S., Jafari, R.: ICA and ANN modeling for photocatalytic removal of
pollution in wastewater. Math. Computat. Appl.22, 38–48 (2017)
7. Razvarz, S., Jafari, R., Gegov, A., Yu, W., Paul, S.: Neural network approach to
solving fully fuzzy nonlinear systems. In: Fuzzy Modeling and Control Methods
Application and Research, pp. 45–68. Nova Science Publisher, Inc., New York
(2018). ISBN: 978-1-53613-415-5
8. Razvarz, S., Jafari, R.: Intelligent techniques for photocatalytic removal of pollu-
tion in wastewater. J. Electr. Eng.5, 321–328 (2017).https://doi.org/10.17265/
2328-2223/2017.06.004
9. Graupe, D.: Statistical training. In: Chen, W., Mlynski, D.A. (eds.) Principles of
Artiﬁcial Neural Networks. Advanced Series in Circuits and Systems, vol. 7, pp.
311–327. World Scientiﬁc (2013)
10. Jafari, R., Yu, W., Li, X.: Solving fuzzy diﬀerential equation with Bernstein neural
networks. In: IEEE International Conference on Systems, Man, and Cybernetics,
Budapest, Hungary, pp. 1245–1250 (2016)
11. Jafari, R. Yu, W.: Uncertain nonlinear system control with fuzzy diﬀerential equa-
tions and Z-numbers. In: 18th IEEE International Conference on Industrial Tech-
nology, Canada, pp. 890–895 (2017).https://doi.org/10.1109/ICIT.2017.7915477
12. Jafarian, A., Measoomy Nia, S., Jafari, R.: Solving fuzzy equations using neural
nets with a new learning algorithm. J. Adv. Comput. Res.3, 33–45 (2012)
13. Werbos, P.J.: Neuro-control and elastic fuzzy logic: capabilities, concepts, and
applications. IEEE Trans. Ind. Electron.40, 170180 (1993)
14. Jafari, R., Yu, W., Razvarz, S., Gegov, A.: Numerical methods for solving fuzzy
equations: a survey. Fuzzy Sets Syst. (2019).https://doi.org/10.1016/j.fss.2019.11.
003. ISSN 0165–0114
15. Kim, J.H., Kim, K.S., Sim, M.S., Han, K.H., Ko, B.S.: An application of fuzzy
logic to control the refrigerant distribution for the multi type air conditioner. In:
Proceedings IEEE International Fuzzy Systems Conference, vol. 3, pp. 1350–1354
(1999)
16. Wakami, N., Araki, S., Nomura, H.: Recent applications of fuzzy logic to home
appliances. In: Proceedings IEEE International Conference Industrial Electronics,
Control, and Instrumentation, Maui, HI, p. 155160 (1993)
17. Jafari, R., Razvarz, S.: Solution of fuzzy diﬀerential equations using fuzzy Sumudu
transforms. In: IEEE International Conference on Innovations in Intelligent Sys-
tems and Applications, pp. 84–89 (2017)
18. Jafari, R., Razvarz, S., Gegov, A., Paul, S.: Fuzzy modeling for uncertain non-
linear systems using fuzzy equations and Z-numbers. In: Advances in Computa-
tional Intelligence Systems: Contributions Presented at the 18th UK Workshop on
Computational Intelligence, 5–7 September 2018, Nottingham, UK. Advances in
Intelligent Systems and Computing, vol. 840, pp. 66–107 (2018)
19. Jafari, R., Razvarz, S.: Solution of fuzzy diﬀerential equations using fuzzy Sumudu
transforms. Math. Comput. Appl., 1–15 (2018)
20. Jafari, R., Razvarz, S., Gegov, A.: Solving diﬀerential equations with z-numbers
by utilizing fuzzy Sumudu transform. In: Intelligent Systems and Applications,
IntelliSys 2018. Advances in Intelligent Systems and Computing, vol. 869, pp.
1125–1138. Springer, Cham (2019)
21. Yu, W., Jafari, R.: Fuzzy modeling and control with fuzzy equations and Z-number.
In: IEEE Press Series on Systems Science and Engineering. Wiley-IEEE Press
(2019). ISBN-13: 978-1119491552
22. Negoita, C.V., Ralescu, D.A.: Applications of Fuzzy Sets to Systems Analysis.
Wiley, New York (1975)

24 R. Jafari et al.
23. Zadeh, L.A.: Probability measures of fuzzy events. J. Math. Anal. Appl.23, 421–
427 (1968)
24. Zadeh, L.A.: Calculus of Fuzzy Restrictions. Fuzzy sets and Their Applications to
CognitiveandDecision Processes, pp. 1–39. Academic Press, New York (1975)
25. Zadeh, L.A.: Fuzzy logic and the calculi of fuzzy rules and fuzzy graphs. Multiple-
Valued Logic1, 1–38 (1996)
26. Razvarz, S., Jafari, R.: Experimental study of Al2O3 nanoﬂuids on the thermal
eﬃciency of curved heat pipe at diﬀerent tilt angle. J. Nanomater., 1–7 (2018)
27. Razvarz, S., Vargas-Jarillo, C., Jafari, R.: Pipeline monitoring architecture based
on observability and controllability analysis. In: IEEE International Conference on
Mechatronics (ICM), Ilmenau, Germany, vol. 1, pp. 420–423 (2019).https://doi.
org/10.1109/ICMECH.2019.872287
28. Razvarz, S., Vargas-jarillo, C., Jafari, R., Gegov, A.: Flow control of ﬂuid in
pipelines using PID controller. IEEE Access7, 25673–25680 (2019).https://doi.
org/10.1109/ACCESS.2019.2897992
29. Razvarz, S., Jafari, R.: Experimental study of Al2O3 nanoﬂuids on the thermal
eﬃciency of curved heat pipe at diﬀerent tilt angle. In: 2nd International Congress
on Technology Engineering and Science, ICONTES, Malaysia (2016)
30. Zadeh, L.A.: A note on Z-numbers. Inform. Sci.181, 29232932 (2011)
31. Jiang, W., Xie, C., Luo, Y., Tang, Y.: Ranking Z-numbers with an improved rank-
ing method for generalized fuzzy numbers. J. Intell. Fuzzy Syst.32, 1931–1943
(2017)
32. Yaakob, A.M., Gegov, A.: Fuzzy rule based approach with Z-numbers for selection
of alternatives using TOPSIS. In: Proceedings of the IEEE International Confer-
ence on Fuzzy Systems, pp. 1–8 (2015)
33. Zamri, N., Ahmad, F., Rose, A.N.M., Makhtar, M.: A fuzzy TOPSIS with Z-
numbers approach for evaluation on accident at the construction site. In: Proceed-
ings of the International Conference on Soft Computing and Data Mining, pp.
41–50 (2016)
34. Azadeh, A., Saberi, M., Atashbar, N.Z., Chang, E., Pazhoheshfar, P.: Z-AHP: a Z-
number extension of fuzzy analytical hierarchy process. In: Proceedings 7th IEEE
International Conference on Digital Ecosystems and Technologies, pp. 141–147
(2013).https://doi.org/10.1109/DEST.2013.6611344
35. Aliev, R.A., Alizadeh, A.V., Huseynov, O.H.: The arithmetic of discrete Z-numbers.
Inform. Sci.290, 134–155 (2015).https://doi.org/10.1016/j.ins.2014.08.024
36. Aliev, R.A., Huseynov, O.H., Zeinalova, L.M.: The arithmetic of continuous Z-
numbers. Inform. Sci.373, 441–460 (2016).https://doi.org/10.1016/j.ins.2016.08.
078
37. De Barros, L.C., Bassanezi, R.C., Lodwick, W.A.: The extension principle of Zadeh
and fuzzy numbers. In: A First Course in Fuzzy Logic, Fuzzy Dynamical Systems,
and Biomathematics. Studies in Fuzziness and Soft Computing, vol. 347, pp. 23–41.
Springer, Heidelberg (2017)
38. Nuriyev, A.M.: Application of Z-numbers based approach to project risks assess-
ment. Eur. J. Interdisc. Stud.5, pp. 67–73 (2019). ISSN 2411-4138
39. Abiyev, R.H., Uyar, K., Ilhan, U., Imanov, E., Abiyeva, E.: Estimation of food
security risk level using Z-number-based fuzzy system. J. Food Q., 9 (2018). Article
ID 2760907.https://doi.org/10.1155/2018/2760907
40. Babanlia, M.B., Huseynov, V.M.: Z-number-based alloy selection problem. In: 12th
International Conference on Application of Fuzzy Systems and Soft Computing,
ICAFS 2016, 29–30 August 2016, Vienna, Austria, Procedia Computer Science,
vol. 102, pp. 183–189 (2016)

Applications of Z-Numbers and Neural Networks in Engineering 25
41. Parrales, A., Colorado, D., Diaz-Gomez, J.A., Huicochea, A., Alvarez, A., Hernan-
dez, J.A.: New void fraction equations for two-phase ﬂow in helical heat exchangers
using artiﬁcial neural networks. Appl. Therm. Eng.130, 149–160 (2018)
42. Alimoradi, H., Shams, M.: Optimization of subcooled ﬂow boiling in a vertical
pipe by using artiﬁcial neural network and multi objective genetic algorithm. Appl.
Therm. Eng.111, 1039–1051 (2017)
43. Chang, W., Chu, X., Fareed, A.F.B.S., Pandey, S., Luo, J., Weigand, B., Laurien,
E.: Heat transfer prediction of supercritical water with artiﬁcial neural networks.
Appl. Therm. Eng.131, 815–824 (2018)
44. Jafari, R., Razvarz, S., Gegov, A.: Neural network approach to solving fuzzy non-
linear equations using Z-numbers. IEEE Trans. Fuzzy Syst. (2019).https://doi.
org/10.1109/TFUZZ.2019.2940919
45. Jafari, R., Yu, W., Li, X., Razvarz, S.: Numerical solution of fuzzy diﬀerential
equations with Z-numbers using bernstein neural networks. Int. J. Comput. Intell.
Syst.10, 1226–1237 (2017)
46. Jafari, R., Yu, W., Li, X.: Numerical solution of fuzzy equations with Z-numbers
using neural networks. Intell. Auto. Soft Comput., 1–7 (2017)

5G-FOG: Freezing of Gait Identiﬁcation
in Multi-class Softmax Neural Network
Exploiting 5G Spectrum
Jan Sher Khan
1
, Ahsen Tahir
2
, Jawad Ahmad
3(B)
, Syed Aziz Shah
4
,
Qammer H. Abbasi
5
, Gordon Russell
3
, and William Buchanan
3
1
University of Gaziantep, Gaziantep, Turkey
2
Glasgow Caledonain University, Glasgow, UK
3
Edinburgh Napier University, Edinburgh, UK
[email protected]
4
Manchester Metropolitan University, Manchester, UK
5
University of Glasgow, Glasgow, UK
Abstract.Freezing of gait (FOG) is one of the most incapacitating and
disconcerting symptom in Parkinson’s disease (PD). FOG is the result of
neural control disorder and motor impairments, which severely impedes
forward locomotion. This paper presents the exploitation of 5G spectrum
operating at 4.8 GHz (a potential Chinese frequency band for Internet
of Things) to detect the freezing episodes experienced by PD patients.
The core idea is to utilize wireless devices such as network interface
card (NIC), radio frequency (RF) signal generator and dipole antennas
to extract the wireless channel characteristics containing the variances
amplitude information that can be integrated into the 5G communica-
tion system. Five diﬀerent human activities were performed including
sitting on chair, slow-walk, fast-walk, voluntary stop and FOG episodes.
A multi-class, multilayer full softmax neural network was trained on the
obtained data for classiﬁcation and performance evaluation of the pro-
posed system. A high classiﬁcation accuracy of 99.3% was achieved for
the aforementioned activities, compared with the existing state-of-the-
art detection systems.
Keywords:Parkinson’s disease
·FOG·Classiﬁcation·Softmax
neural network
1 Introduction
Parkinson’s disease (PD) is a progressive neurodegenerative disease described
by Parkinson in 1817 [1]. Over time, PD eﬀectively progresses and worsen and
hence called a progressive disease. A speciﬁc type of neuron known as dopamine
neuron losses during PD that causes FOG. FOG is a serious gait disorder which
interrupts walking with a transient and sudden nature. Due to sudden and seri-
ous debilitating nature, FOG disturbs the balance of PD patients and therefore
cﬀSpringer Nature Switzerland AG 2020
K. Arai et al. (Eds.): SAI 2020, AISC 1230, pp. 26–36, 2020.
https://doi.org/10.1007/978-3-030-52243-8
_3

5G-FOG 27
causes falls that may lead to mortality [2,3]. The pathophysiology of FOG is still
under research and its treatment is a still an open clinical challenge [4]. However,
recently, authors in [5] reported the impact of levodopa-carbidopa intestinal gel
(LCIG) FOG and concluded that a long term control over FOG is possible via
LCIG if FOG is detected correctly. Furthermore, it is suggested in [5]thatanum-
ber of experiments are required with correct identiﬁcation of FOG in patients.
Authors in [6] reduced FOG and improved mobility via simultaneously tar-
geting motor and cognitive regions through transcranial direct current stimula-
tion. Though, authors [6] reported the improvement of mobility but correctly
predicting the state of freezing was overlooked. Therefore, to decrease the fall
rate and before providing a solution for FOG, a system must be developed to
detect FOG with higher accuracy. FOG can detected through numerous detec-
tion systems such as wearable devices and camera etc [7–11]. However, there are
several limitations associated with camera-based and wearable based systems.
For instance, the camera-based system works raise privacy concerns due to the
constant recording of images or videos. In addition, they are computationally
expensive as well since processing images or videos require dedicated hardware.
On the other hand, wearable devices have to be worn by the subject’s all the
time due to which the patients might feel uncomfortable. Moreover, more often
than not, the patients forget to wear the devices after changing clothes or taking
a shower. Due to aforementioned issues, it is evident that other digital medium
should be investigated. This paper presents a wireless channel information (WCI)
based new detection method. A device free wireless sensing method is developed
and the accuracy of the proposed scheme is tested using artiﬁcial neural network
(ANN).
Over the past few years, ANN has been applied in a number of areas including
speech recognition [12], image classiﬁcation [13], and energy demand prediction.
Rahim et al. [12] and Chu et al. [14] applied ANN to the speech recognition.
Moreover, ANN-based algorithms have also been used in image classiﬁcation and
recognition [13,15,16]. Previously, Neural network based schemes are applied to
chemical-related research, molecular biology, medicines, environmental sciences
and ecosystems [17–20]. This paper exploits the application of multi-class, full
softmax multilayer feedforward neural network (ML-FFNN) using WCI and 5G
spectrum for FOG detection.
The core idea of the proposed work is to detect the FOG episode by clas-
sifying various human activities such as sitting/standing on chair, slow-walk,
fast-walk, voluntary stop. The classiﬁcation performed using variations in WCI
data is received through wireless devices including RF signal generator, networks
interface card (NIC) and dipole antenna [21–24].
2 Experimental Setup
The general experimental setup for FOG detection is shown in Fig.1. The exper-
iment was conducted in a room with dimensions (15 m×15 m) in New Science
Building, Xidian University, China. The experimental settings included an RF

28 J. S. Khan et al.
generator (DSG3000 Series), two dipole antennas, TP-link (PCE-AC68) next
generation dual-band wireless AC1900 PCIe adapter NIC, and HP desktop com-
puter with Ubuntu 14.10 (64 bits) and 4 GB RAM. The RF signal generator
connected with the dipole antenna operating at 4.8 GHz was set as an Access
Point (AP) to generate RF signals at multiple frequencies. The network interface
card wired with dipole antenna embedded in a desktop computer received the
seamless WCI data. The transmitter and receiver were kept 10 m apart from
each other.
A total number of 15 volunteers took part in the experimental campaign and
were asked to perform the aforementioned ﬁve activities. Each human activity
constantly disturbed the wireless medium and the unique WCI imprint induced
was used for activity recognition.
RF generator
Object under
Observa?on
Receiving antenna and PC
for observa?on analyzing
Fig. 1.General setup of the experiment.
3 FOG Detection Methodology
The design of the FOG system is presented in Fig.2which consists of three main
parts:
1). Wireless channel information
2). Feature extraction
3). Multi-class softmax ML-FFNN training & classiﬁcation for FOG detection
Step 1:Exploiting the IEEE 802.11n standards for orthogonal frequency divi-
sion multiplexing (OFDM), which divide a single channel carrier into several
subcarriers and enables the data to be transmitted in parallel to solve multipath
fading problem [25]. The signal received using network interface card can be
computed as:
Y=(H×X)+N (1)

5G-FOG 29
Fig. 2.Flowchart of the proposed FOG detection methodology.
HereXandYare the transmitted and received signals, respectively.Ndenotes
the channel noise whileHdemonstrates channel frequency response (CFR) of
the wireless channel data which is a complex number.
H=[h
1,h2,h3, ......,h n] (2)
h
n=φh nφexp
j∠h n
(3)
In Eq.3,φh
nφrepresents the variance of amplitude information and∠h n
describes phase information fornsub-carrier. It should be noted that the phase
information obtained via NIC is extremely random and cannot be used for any
application. Therefore, in this paper we have used the variances of amplitude
information for training and testing the ANN algorithm to classify FOG from
other daily life activities in an accurate and eﬃcient way.
Step 2:In this step, time domain features such as mean, standard deviation,
skewness, kurtosis, mean absolute deviation (MAD), interquartile range (IQR)
and peaks are extracted from the WCI data and plugged into the levenberg-
marquardt (LM) training algorithm. Features extraction is primarily data reduc-
tion by ﬁnding the most informative variables-based subset of the same dataset.
Mean is deﬁned as the average of all data points in a data matrix and specify
the variability around a distinct value in some data matrix. Mean can be more
eﬀective in case of relatively uniformly spread data with no extraordinarily high
or low values. Mathematically, mean is deﬁned as:
μ
x=
φ
a×b
i=1
xi
a×b
, (4)

30 J. S. Khan et al.
(a) (b)
(c) (d)
(e)
Fig. 3.Perturbations of amplitude information of 30 subcarriers. (a) Walking slow. (b)
Walking fast. (c) Sit-stand on chair. (d) Voluntary stop. (e) FOG.

5G-FOG 31
whereaandbrepresent the number of rows and columns of a data matrix,
respectively.x
iis a data point at indexi. Standard deviation is known as the
spread (variability) of data points in a data matrix. Mathematically standard
deviationsd
Xcanbemeasuredas[26]:
sd
x=
1
a×b
a×b
∞
i=1
(xi−μx)
2
(5)
Information about the spread of data can also be obtained via interquartile
range [26].q
ucomputes the middle value above the median, whileq lcomputes
the middle value below the median of a data set. Mathematically interquartile
range can be written as:
iqr=q
u−ql (6)
Skewnesss
xcomputes the asymmetry of the probability distribution while kur-
tosisk
xcomputes the shape of the probability distribution of a real-valued ran-
dom variable. Skewnesss
xand kurtosisk xcan be used to make judgments about
image surfaces. Mathematically skewness and kurtosis can be computed as [27]:
s
x=
1
a×b
×
a×b
∞
i=1
(
x
i−μx
sdx
)
3
(7)
k
x=
1
a×b
×
a×b
∞
i=1
(
x
i−μx
sdx
)
4
(8)
The mean absolute deviation about mean measure the dispersion of X about its
mean and can be mathematically written as [28]:
mad
x=
φ
a×b
i=1
|xi−μx|
a×b
(9)
Step 3:Due to the faster operations, smaller training datasets requirement,
easy implementation and ability to learn quickly, we have utilized a multi-layer
perceptron neural network (MLPNN) with a single input layer, single hidden and
single output layer as shown in Fig.4. Levenberg-Marquardt (LM) [29] training
algorithm is used during feature training process. LM is an iterative method
that is used for solving non-linear minimization problem. The proposed classiﬁer
identify FOG episodes which is distinguishable from other routine activities using
the proposed method. The input layer consists of seven neurons while the output
layer consists of ﬁve neurons since we are classifying ﬁve diﬀerent activities.
Sigmoid activation function is used for input and output layers. Hidden layer
which consist of ten neurons uses linear softmax activation function. Sigmoid
function maps the interval (−∞,∞)onto(0,1) while softmax function squashes
anxsize vector between 0 and 1. Furthermore, softmax function normalized
the exponential function to make the sum of whole vector equal to 1. Therefore,
the output softmax function interpret a set of speciﬁc features belong to certain

32 J. S. Khan et al.
Fig. 4.MLPNN schematic diagram.
class. Mathematically sigmoid (φ) and softmax (Φ) functions can be computed
as in [30] and in [31], respectively:
φ(x)=
1
1 + exp
−x
(10)
Φ(x
i)=
exp
xi
φ
N
n
exp
xn
(11)
4 Result and Discussion
The variances of amplitude information for 30 subcarriers obtained using wireless
devices exploiting 5G spectrum of ﬁve diﬀerent activities is presented in Fig.3,
respectively. In Fig.3, x-axis indicate total number of subcarriers, y-axis shows
the total number of packets and z-axis is the relative power in dB indicating the
variations in amplitude information. It can be observed that each human activity
has resulted in a unique WCI signature which can be classiﬁed using multi-
class ML-FFNN with the LM learning algorithm. Figure5shows the overall
time history of ﬁve human activities for subcarrier number 6
th
. It illustrates the
relative power level ﬂuctuated between 4 dB to 16 dB for packet number 1 to
220. However, there is a shift in power level when the subject stands stationary
(with small scale body movements such as breathing or small limb movements).
Moreover, the power level varied around 24 dB when person was asked to walk
slowly within area of interest. An increase in the variances of power level in

5G-FOG 33
Fig. 5.Amplitude variation of a random subcarrier.
packet numbers 900 to 1800 is observed when the person walks at a fast pace.
While, for FOG episodes, the variations power level ﬂuctuations between 24 dB
to 26 dB are observed.
Table1illustrates the performance of our system as compared to the state-
of-the-art latest works [7–11,32–36] in the domain of FOG detection leveraging
traditional systems, such as wearable devices, smart phone sensors and vision
based systems. The proposed system exploits 5G spectrum to detect and classify
FOG with a high accuracy of 99.3% (see confusion matrix, Table2) with an
increase of approximately 6% over the second best method [33].
Table 1.Comparison of FOG detection systems
Authors Detection system Types of sensors Algorithm Accuracy
Prateek et al. [7] Wearable devices Inertial measurement
unit
Generalized likelihood ratio test (GLRT) 81.03%
Camps et al. [8] Wearable devices Inertial measurement
unit
Convolution neural network (CNN) 89%
Sam`aetal.[9] Wearable devices Accelerometer Support vector
machine
89.6%
Rodr´ıguez et al. [32]Wearable devices Accelerometer Support vector
machine
76.8%
Aminis et al. [11] Vision based Camera, depth sensor position/head oﬀset &
angle tracking
86.6%*
Bigy et al. [10] Vision based Camera, depth sensor subject/body joint
positions
92%
Kim et al. [33] Smart phone Accelerometer,
gyroscope
Convolution neural network (CNN) 93.8%
Capecci et al. [34] Smart phone Accelerometer Power spectrum and
cadence measures
92.86%
Kim et al. [35] Smart phone Accelerometer,
gyroscope
AdaBoost.M1 86%
Pepa et al. [36] Smart phone Accelerometer Fuzzy inference
system
89%
Proposed 5G spectrum Wireless sensing Multi-class softmax
FFNN
99.3%

34 J. S. Khan et al.
Table 2.Confusion matrix
Output Class 1
231
10.5%
0
0.0%
0
0.0%
1
0.0%
0
0.0%
99.6%
0.4%
2
2
0.1%
285
13.0%
0
0.0%
3
0.1%
0
0.0%
98.3%
1.7%
3
2
0.1%
0
0.0%
200
9.1%
1
0.0%
0
0.0%
98.5%
1.5%
4
0
0.0%
3
0.1%
0
0.0%
976
44.4%
1
0.0%
99.6%
0.4%
5
0
0.0%
0
0.0%
0
0.0%
3
0.1%
492
22.4%
99.4%
0.6%
98.3%
1.7%
99.0%
1.0%
100%
0.0%
99.2%
0.8%
99.8%
0.2%
99.3%
0.7%
1 234 5
Target Class
5 Conclusion
This study presented the design and implementation of an FOG system lever-
aging wireless devices operating at 4.8 GHz (compatible with 5G spectrum for
IoTs) in conjunction with multi-class softmax feedforward neural networks. The
wireless channel information was extracted for ﬁve diﬀerent human activities in
indoor settings to classify the FOG episodes from sitting on chair, walking slowly,
walking with fact pace and voluntary stop. The multi-class ML-FFNN leverag-
ing features such as mean, standard deviation, skewness, kurtosis and peaks of
power spectrum were used to classify the particular human activities. It was
observed that the system provided an average accuracy of 99.3% for various
subjects under test.
References
1. Parkinson, J.: An essay on the shaking palsy (printed by whittingham and rowland
for sherwood, neely, and jones) (1817)
2. Bloem, B.R., Hausdorﬀ, J.M., Visser, J.E., Giladi, N.: Falls and freezing of gait
in Parkinson’s disease: a review of two interconnected, episodic phenomena. Mov.
Disord. Oﬃcial J. Mov. Disord. Soc.19(8), 871–884 (2004)
3. Canning, C.G., Paul, S.S., Nieuwboer, A.: Prevention of falls in Parkinson’s disease:
a review of fall risk factors and the role of physical interventions. Neurodegener.
Dis. Manage.4(3), 203–221 (2014)
4. Amboni, M., Barone, P., Picillo, M., Cozzolino, A., Longo, K., Erro, R., Iavarone,
A.: A two-year follow-up study of executive dysfunctions in Parkinsonian patients
with freezing of gait at on-state. Mov. Disord.25(6), 800–802 (2010)
5. Zibetti, M., Angrisano, S., Dematteis, F., Artusi, C.A., Romagnolo, A., Merola, A.,
Lopiano, L.: Eﬀects of intestinal levodopa infusion on freezing of gait in parkinson
disease. J. Neurol. Sci.385, 105–108 (2018)

5G-FOG 35
6. Dagan, M., Herman, T., Harrison, R., Zhou, J., Giladi, N., Ruﬃni, G., Manor, B.,
Hausdorﬀ, J.M.: Multitarget transcranial direct current stimulation for freezing of
gait in Parkinson’s disease. Mov. Disord.33(4), 642–646 (2018)
7. Prateek, G.V., Skog, I., McNeely, M.E., Duncan, R.P., Earhart, G.M., Nehorai,
A.: Modeling, detecting, and tracking freezing of gait in Parkinson disease using
inertial sensors. IEEE Trans. Biomed. Eng.65(10), 2152–2161 (2018)
8. Camps, J., Sam`a, A., Mart´ın, M., Rodr´ıguez-Mart´ın, D., P´erez-L´opez, C., Moreno
Arostegui, J.M., Cabestany, J., Catal`a, A., Alcaine, S., Mestre, B., Prats, A.,
Crespo-Maraver, M.C., Counihan, T.J., Browne, P., Quinlan, L.R., Laighin, G.,
Sweeney, D., Lewy, H., Vainstein, G., Costa, A., Annicchiarico, R., Bay´es,
`
A.,
Rodr´ıguez-Molinero, A.: Deep learning for freezing of gait detection in Parkin-
son’s disease patients in their homes using a waist-worn inertial measurement unit.
Knowl.-Based Syst.139, 119–131 (2018)
9. Sam`a, A., Rodr`ıguez-Mart`ın, D., P´erez-L´opez, C., Catal`a, A., Alcaine, S., Mestre,
B., Prats, A., Crespo, M.C., Bay´es,
`
A.: Determining the optimal features in freez-
ing of gait detection through a single waist accelerometer in home environments.
Pattern Recogn. Lett.105, 135–143 (2018)
10. Bigy, A.A.M., Banitsas, K., Badii, A., Cosmas, J.: Recognition of postures and
Freezing of Gait in Parkinson’s disease patients using Microsoft Kinect sensor. In:
2015 7th International IEEE/EMBS Conference on Neural Engineering (NER),
pp. 731–734. IEEE (2015)
11. Amini, A., Banitsas, K., Young, W.R.: Kinect4fog: monitoring and improv-
ing mobility in people with parkinson’s using a novel system incorporating the
Microsoft Kinect v2. Disabil. Rehabil. Assist. Technol. 1–8 (2018)
12. Rahim, M.G., Goodyear, C.C., Kleijn, W.B., Schroeter, J., Sondhi, M.M.: On the
use of neural networks in articulatory speech synthesis. J. Acoust. Soc. Am.93(2),
1109–1121 (1993)
13. Kung, S.-Y., Taur, J.-S.: Decision-based neural networks with signal/image classi-
ﬁcation applications. IEEE Trans. Neural Netw.6(1), 170–181 (1995)
14. Chu, W., Bose, N.: Speech signal prediction using feedforward neural network.
Electron. Lett.34(10), 999–1001 (1998)
15. DeKruger, D., Hunt, B.R.: Image processing and neural networks for recognition
of cartographic area features. Pattern Recogn.27(4), 461–483 (1994)
16. Cosatto, E., Graf, H.P.: A neural network accelerator for image analysis. IEEE
Micro3, 32–38 (1995)
17. Kvasniˇcka, V.: An application of neural networks in chemistry. Chem. Pap.44(6),
775–792 (1990)
18. Lerner, B., Levinstein, M., Rosenberg, B., Guterman, H., Dinstein, L., Romem,
Y.: Feature selection and chromosome classiﬁcation using a multilayer perceptron
neural network. In: 1994 IEEE International Conference on Neural Networks. IEEE
World Congress on Computational Intelligence (1994)
19. Lek, S., Gu´egan, J.-F.: Artiﬁcial neural networks as a tool in ecological modelling,
an introduction. Ecol. Model.120(2–3), 65–73 (1999)
20. Colasanti, R.: Discussions of the possible use of neural network algorithms in eco-
logical modeling. Binary Comput. Microbiol.3(1), 13–15 (1991)
21. Yang, X., Shah, S.A., Ren, A., Zhao, N., Zhao, J., Hu, F., Zhang, Z., Zhao, W.,
Rehman, M.U., Alomainy, A.: Monitoring of patients suﬀering from REM sleep
behavior disorder. IEEE J. Electromagn. RF Microwaves Med. Biol.2(2), 138–143
(2018)

36 J. S. Khan et al.
22.Yang,X.,Shah,S.A.,Ren,A.,Zhao,N.,Fan,D.,Hu,F.,Ur-Rehman,M.,von
Deneen, K.M., Tian, J.: Wandering pattern sensing at s-band. IEEE J. Biomed.
Health Inform. (2017)
23. Wang, X., Yang, C., Mao, S.: Tensorbeat: Tensor decomposition for monitoring
multiperson breathing beats with commodity wiﬁ. ACM Trans. Intell. Syst. Tech-
nol. (TIST)9(1), 8 (2017)
24. Zhang, D., Wang, H., Wu, D.: Toward centimeter-scale human activity sensing
with wi-ﬁ signals. Computer50(1), 48–57 (2017)
25. Cao, W., Li, X., Hu, W., Lei, J., Zhang, W.: Ofdm reference signal reconstruc-
tion exploiting subcarrier-grouping-based multi-level lloyd-max algorithm in pas-
sive radar systems. IET Radar Sonar Navig.11(5), 873–879 (2016)
26. Kreyszig, E.: Advanced Engineering Mathematics. Wiley, Hoboken (2010)
27. Kumar, V., Gupta, P.: Importance of statistical measures in digital image process-
ing. Int. J. Emerg. Technol. Adv. Eng.2(8), 56–62 (2012)
28. El Amir, E.A.H.: On uses of mean absolute deviation: decomposition, skewness
and correlation coeﬃcients. Metron70(2–3), 145–164 (2012)
29. Hagan, M.T., Demuth, H.B., Beale, M.H., De Jes´us, O.: Neural Network Design,
vol. 20. Pws Pub., Boston (1996)
30. Karlik, B., Olgac, A.V.: Performance analysis of various activation functions in
generalized mlp architectures of neural networks. Int. J. Artif. Intell. Expert Syst.
1(4), 111–122 (2011)
31. Tang, Y.: Deep learning using linear support vector machines. arXiv preprint
arXiv:1306.0239(2013)
32. Rodr´ıguez-Mart´ın, D., Sam`a, A., P´erez-L´opez, C., Catal`a, A., Moreno Arostegui,
J.M., Cabestany, J., Bay´es,
`
A., Alcaine, S., Mestre, B., Prats, A., Crespo, M.C.,
Counihan, T.J., Browne, P., Quinlan, L.R.,
´
OLaighin, G., Sweeney, D., Lewy,
H., Azuri, J., Vainstein, G., Annicchiarico, R., Costa, A., Rodr´ıguez-Molinero,
A.: Home detection of freezing of gait using support vector machines through
a single waist-worn triaxial accelerometer. PLoS ONE12(2), e0171764 (2017).
http://dx.plos.org/10.1371/journal.pone.0171764
33. Kim, H.B., Lee, H.J., Lee, W.W., Kim, S.K., Jeon, H.S., Park, H.Y., Shin,
C.W., Yi, W.J., Jeon, B. and Park, K.S.: Validation of freezing-of-gait mon-
itoring using smartphone. Telemed. e-Health24(12) (2018). tmj.2017.0215.
https://www.liebertpub.com/doi/10.1089/tmj.2017.0215
34. Capecci, M., Pepa, L., Verdini, F., Ceravolo, M.G.: A smartphone-based architec-
ture to detect and quantify freezing of gait in Parkinson’s disease. Gait Posture
50, 28–33 (2016).http://dx.doi.org/10.1016/j.gaitpost.2016.08.018
35. Kim, H., Lee, H.J., Lee, W., Kwon, S., Kim, S.K., Jeon, H.S., Park, H., Shin,
C.W., Yi, W.J., Jeon, B.S., Park, K.S.: Unconstrained detection of freezing of Gait
in Parkinson’s disease patients using smartphone. In: Proceedings of the Annual
International Conference of the IEEE Engineering in Medicine and Biology Society,
EMBS, November 2015, pp. 3751–3754 (2015)
36. Pepa, L., Ciabattoni, L., Verdini, F., Capecci, M., Ceravolo, M.G.: Smartphone
based Fuzzy Logic freezing of gait detection in Parkinson’s Disease. In: MESA
2014 - 10th IEEE/ASME International Conference on Mechatronic and Embedded
Systems and Applications, Conference Proceedings (2014)

Adaptive Blending Units: Trainable
Activation Functions for Deep Neural
Networks
Leon Ren´eS¨utfeld
1(B)
, Flemming Brieger
2
, Holger Finger
1
,SonjaF¨ullhase
1
,
and Gordon Pipa
1
1
Institute for Cognitive Science, Wachsbleiche 27, 49090 Osnabr¨uck, Germany
[email protected]
2
Ulm University, Helmholtzstraße 16, 89081 Ulm, Germany
[email protected]
Abstract.The most widely used activation functions in current deep
feed-forward neural networks are rectiﬁed linear units (ReLU), and many
alternatives have been successfully applied, as well. However, none of the
alternatives have managed to consistently outperform the rest and there
is no uniﬁed theory connecting properties of the task and network with
properties of activation functions for most eﬃcient training. A possi-
ble solution is to have the network learn its preferred activation func-
tions. In this work, we introduce Adaptive Blending Units (ABUs), a
trainable linear combination of a set of activation functions. Since ABUs
learn the shape, as well as the overall scaling of the activation function,
we also analyze the eﬀects of adaptive scaling in common activation
functions. We experimentally demonstrate advantages of both adaptive
scaling and ABUs over common activation functions across a set of sys-
tematically varied network speciﬁcations. We further show that adaptive
scaling works by mitigating covariate shifts during training, and that the
observed advantages in performance of ABUs likewise rely largely on the
activation function’s ability to adapt over the course of training.
Keywords:Adaptive Blending Units
∙Trainable∙Activation
functions
∙Deep learning∙Convolutional networks
1 Introduction
Deep neural networks are structured around layers, each of which performs a
linear transformation of its input before feeding the signal through a scalar non-
linear activation function. Chaining larger numbers of non-linear functions then
allows the networks to ﬁnd and extract complex features in the input. Activation
functions thus have a key function in deep neural networks: Without intermit-
tent non-linearities, these networks could only perform linear operations on the
input. But despite a large number of activation functions proven successful in the
literature, it remains unclear, what properties of an activation function are most
cﬀSpringer Nature Switzerland AG 2020
K. Arai et al. (Eds.): SAI 2020, AISC 1230, pp. 37–50, 2020.
https://doi.org/10.1007/978-3-030-52243-8
_4

38 L. R. S¨utfeld et al.
desirable, given a particular task and network conﬁguration. Ideally, the network
would sort this issue out by itself, but most common activation functions are
ﬁxed during training, i.e., their shape and scaling are treated as hyperparame-
ters. We suggest changing this practice by making an activation function’s shape
and scaling a trainable parameter of the network. Our main contribution in this
work is the Adaptive Blending Unit (ABU), a linear combination of a set of basic
activation functions that allows the shape and scaling of the resulting activation
function to be learned during training. In an eﬀort to separate and understand
the eﬀects of the activation function’s shape and its scaling, we also examine the
eﬀect of adaptive scaling on common activation functions without adaptation of
the shape, as well as normalizing the blending weights in ABUs, thus learning its
shape without learning any scaling. Throughout this work, we apply one scaling
weight, or one set of blending weights (i.e., one ABU) per layer of the network.
This way, the network is free to learn the activation function and/or scaling that
best suits the computations performed in any given layer, while the number of
parameters in the network is kept low enough, as not to require regularization.
The remainder of this work is structured as follows. In Sect.2, we will review
related work, before comparing and analyzing common activation functions, their
adaptively scaled counterparts and ABUs on CIFAR image classiﬁcation tasks
in Sect.3. In Sect.4, we examine multiple ways of normalizing ABUs, to provide
an account of adaptive shape without adaptive scaling. Finally, in Sect.5,we
examine pre-training of the scaling and blending weights, to examine the role of
adaptiveness over the course of training. We conclude the paper by discussing
limitations of the chosen approach, and providing an outlook on future work on
this topic.
2 Related Work
The most prevalent activation function in modern neural networks is theRecti-
ﬁed Linear Unit (ReLU)[10,21], a piecewise-linear function returning the iden-
tity for all positive inputs and zero otherwise. Its constant derivative of 1 on
the positive part helps alleviating the vanishing gradient problem [7], making it
the ﬁrst activation function allowing for a large number of stacked layers to be
trained eﬃciently. With this, ReLU was partly responsible for the breakthrough
of deep neural networks around 2012, marked by AlexNet’s victory in the annual
ILSVRC challenge [16].Leaky ReLU (LReLU)[19],Parametric ReLU (PReLU)
[11], andRandomized Leaky ReLU (RReLU)[25] are all based on ReLU, but
replace the zero-output for negative values by a linear function. In PReLU, the
slope of the negative part of the function is controlled by a trainable parameter.
Exponential Linear Units (ELU) [4] like ReLU, return the identity for positive
values, butα(exp(x)−1) for negative values, withαtypically set to 1. Scaled
Exponential Linear Units (SELU) [14] are identical to ELU, except for an addi-
tional scaling parameterλacting upon the function as a whole. The values for
αandλin SELUs are analytically derived to ensure convergence of activations
towards unit mean and variance across layers. In a more empirical approach,

Exploring the Variety of Random
Documents with Different Content

fierce rage and longing, that he was a hundredfold more precious
now?
There are women whom it is very dangerous to love, as in Eden
there stood a tree that it was death to taste. But the forbidden fruit
was gathered nevertheless; and these beauties seem to allure more
than their share of victims, to win more than their natural meed of
triumph. Perhaps it is their destiny to avenge on mankind the
common wrongs of their sex, and to fall at last by the very weapons
they have wielded so successfully in their march over a host of slain.
The old king's eyes were dim, and his senses failed him perceptibly,
as life waned gradually, yet surely, like an unfed lamp, or a leaking
vessel of wine. The pomp of royalty, the joy of battle, the feast, the
pageant, the bright steel quivering in his grasp, the good horse
bounding between his knees, what were they all now but shadows,
memories, vague, idle dreams of the past? Was this the hand, he
was fain to ask himself, that drew the heaviest bow in the broad land
of Shinar, the arm that could drive a javelin through and through the
lion's heart?
Yonder upon the wall was sculptured many a deed of prowess, many
a noble triumph of warfare or the chase. Warriors in long array were
marching to the battle or the siege; archers bent their bows, slingers
and spearmen smote and slew and spared not; horsemen galloped,
chariots rolled, and vultures soared over heaps of corpses. A bank
was raised against a city, the battering-ram laid to its gates, while
amidst a shower of arrows and javelins men were falling headlong
from its walls to feed the fishes in the river below.
Again, linked in a cruel chain, the line of captives paced slowly by,
bearing on their shoulders children, household stuff and goods,
equally the spoil of their conqueror. The men marched sullenly, with
downcast looks; the women beat their breasts and tore their hair.
Here, with hook in his victim's nostrils, or knife to flay his naked
flesh, a fierce warrior tortured some poor suppliant slave. There,
proffering for a tribute the productions of his country—garments,

gold, grain, animals wild and tame—some cringing wretch implored
mercy at the feet of his executioner. But amongst all these scenes of
strife, glory, and rapine, one figure still predominated, tall, fierce,
and stately, the high tiara bound about its brows, bow and spear in
hand; but, whether careering in the war-chariot over prostrate
enemies, or sitting on the throne of state under the royal parasol,
there was still poised above its head the winged mystery within a
circle that heralded the sacred person of a king.
Could this be the same Ninus, he asked himself, whose limbs, so stiff
and aching, now endured his silken robes with less patience than
once they had carried his iron harness, whose head wavered and
nodded on the lean neck that was once a tower of strength, proud,
erect, colossal, like a column of stone?
And that winged figure in the circle. What was it? Did it really hover
over them to protect the race of Nimrod in battle, or was this too a
myth, a fable, a mere imposition of the priests? Should he know
when he went to join his ancestors? and would it be long—how long!
—ere he took his place among the stars?
There was not much to leave, after all! The wild bull had been driven
from the plains, and could be found in no nearer fastness than the
northern mountains now. He had himself exterminated the lion
within the paradise round his palace, and it was weary work to ride
in search of him over the scorching desert. Even the rush of battle
was not what it used to be. Where were the men of the olden time,
such as the champion he slew in Bactria, who stood two palms'
breadths higher than the tallest warrior of either host, leaning on
their spears to witness the single combat between a giant and a
king? Or that fierce Ethiopian in the first Egyptian campaign, whom
Pharaoh's chief counsellor had made captain of his armies for his
matchless valour, and whose sturdy assault caused Ninus to reel and
stagger where he stood, ere the swarthy swordsman went down
under the buffets of the Great King, then in the vigour of his prime?
But in his last expedition the armies of Egypt seemed to give way
without a struggle before his spear, and it was hardly worth while to

bid his chariot driver turn his hand into the press of battle. Even the
wine of Eshcol was tasteless now; the wine of Damascus worse, and
the feast had become loathsome to him as the fray. He was weary of
it all, could give it up without a regret, but for the queen.
Feeling, in spite of his angry protest against his own misgivings, that
the link which bound them together grew slighter every day—that,
like a frayed bowstring, it must snap at last, and leave her free,—the
love in his fierce old heart began to be tinged with a savage and
unreasoning jealousy, such as made him intolerant of every glance
she directed at another, of every moment she was absent from his
side. He had summoned her to his presence with all those forms and
observances, the necessary ceremonial of royalty, which chafed him
now more than ever; and in his impatience he bade the light-footed
Sethos hurry to and fro to see if the queen and her train of
attendants were not yet at the gates, although from where he sat in
his throne of state he could command a noble approach, some
furlongs in length, through double lines of colossal monsters, leading
to the wide entrance of his palace.
A jewelled cup, filled to the brim, stood neglected at his hand. Ever
and anon he stormed at Sethos because the wine had lost its flavour,
and the queen tarried so long.
"I could put on and prove ten suits of harness," said the angry old
monarch, "in less time than it takes a woman to tire her head! And
yet one hair of that comely head is surely better worth preserving
than the whole of this worn-out body of mine, that hath scarce
strength left to draw a bow or empty a cup. Saw you not, Sethos,
how fair she looked on the wall above us when we rode in, slender
and pliant like a spear bending beneath a truss of forage? Who was
attending her, boy? My memory halts and fails me now worse than a
ham-strung steed."
"Kalmim, my lord," answered the cup-bearer, "with certain of the
women, and Sarchedon."

He was too good a courtier to mention Assarac, dreading the storm
a priest's name was likely to bring down in the king's present mood.
"Sarchedon," repeated Ninus—"one of my own guards. A stout
warrior enough, in the boy's play we call fighting now, and a comely
youth—ruddy and comely as a maid. How came he absent from his
duty in the ranks?"
"He had been sent by my lord from the host with the Great King's
signet to the queen," was the reply. "He has remained in attendance
on her ever since."
The old face turned gray with some hidden pang, and the blood-shot
eyes rolled savage under their shaggy brows.
"By the beard of Nimrod, I will take better order with these golden
guards of mine!" exclaimed the king. "Do they think, because
Pharaoh and his bowmen are no longer flying before my chariot, I
have beaten my sword into a pruning-hook, and have forgotten how
to mount a war-chariot or set a company in array? Where is this
deserter now?"
"He is on duty at the great entrance," was the respectful answer.
"My lord the king may see him from where he sits."
Sarchedon, in truth, with a handful of his comrades, was on guard at
the palace gate, conspicuous even amongst those goodly warriors by
the beauty of his person and the splendour of his attire.
Ere the king could summon him to his presence, his attention was
diverted by the approach of his wife, followed by the women of her
household; a fair and fragrant company, that wound through the
avenues of winged bulls and colossal monsters, like a growth of wild
flowers trailing across the surface of a rock.
The king's eyes were not too dim to mark every movement of the
woman he loved. His old heart began to beat faster and the blood
stirred in his veins.

How fair and noble was the bearing of that shapely figure, as it
glided on with the measured step that became her so well! How
delicate and beautiful the pale face! so easily recognised even at a
distance from which its features could not be distinguished, and
bringing back to him as it was unveiled now, on entering her
husband's dwelling, that well-remembered morning in Bactria, when
she rode into the camp serene and radiant, like a star dropped down
from heaven.
What was this? He started, and half rose from his throne; for she
had paused amongst the guards, and one of them had fallen on his
face at her feet.
Semiramis, who was above all the forms and ceremonies that
trammelled weaker natures, breaking through them at will in court,
camp, or palace, had resolved to take signal vengeance on
Sarchedon whenever she should see him, careless alike whether
they met in the desert, on the house-top, or here in the formidable
presence of the king. She knew how to stab him too, and
determined, at whatever cost to her own feelings, she would drive
her thrust home.
How beautiful he looked, standing there in his golden helmet, with
the scarlet-bordered mantle falling from his shoulders, and the white
tunic reaching to his knee! Not Menon, she thought, when he wooed
her by the silver lake that mirrored the towers of Ascalon, was half
so fair; but Menon loved her dearly, while this man—well, she would
make him eat the hardest morsel, drink the bitterest waters of
affliction, and afterward he should die. What would be left her then?
The love of this old dotard, the hollow pageantry, the empty
pleasures, the heavy magnificence of a court. How she loathed them
all! And what good would it do her even to attain supreme power if
she must rule alone, without companionship, without sympathy,
without love?
She had wavered in her purpose a hundred times ere she stepped as
many paces. She was inflexible when she bade Sarchedon come

forward from the line of his comrades, irresolute while he advanced
and pitiless once more as he prostrated himself at her feet.
"You are entitled to ask a request," said she, very coldly and
haughtily, "as having borne hither the signet of my lord the king. It
is my part to intercede with him in your favour, and the old custom
in our land of Shinar bids him grant your desire, even to the half of
his kingdom."
His eyes lightened with pleasure, and her heart turned to stone. Yet
even in that moment she marked that he still wore her amulet round
his neck.
The name of Ishtar was on his lips, but some instinct of the palace—
it may be something in the queen's face—forbade him to pronounce
it. He had wit enough to bow his forehead in the dust, and to
answer,
"I do but desire the light of her countenance, and permission to
abide in the service of the Great Queen."
She was not deceived by his submission, though her eyes shone
with a softer lustre while she continued, "Is there no treasure you
covet, no post of honour you desire, no maiden in the whole land of
Shinar you would fain take home with you to your tent?"
"I may not lift mine eyes to Ashtaroth," was his cautious reply. "If I
must needs choose from among the flowers of earth, I would beg of
the Great Queen to give me Ishtar, the daughter of Arbaces."
She was ready with her blow. Looking him full in the face, with the
calm pitiless smile of one who puts some wounded reptile out of
pain—
"It is too late," she said, in hard cutting accents. "The damsel has
been promised to my son. Even now the prince is lifting her veil to
salute his bride!"

In his agony he fell forward, grasping the queen's robe wildly in his
hand.
The Great King sprang to his feet, his beard bristling, his very
eyebrows shaking with ungovernable anger. For a space he could not
even find voice to speak. Then he burst out,
"By the blood of Nisroch, it is too much! He has laid hands on the
queen before my very face! Were he flesh of my loins and bone of
my body, he should be consumed to ashes. Ho, guards, away with
him! Cover his face and lead him forth!"
A score of hands grasped the offender, a score of spears were
pointed at his breast. Though it was her own act, nay, because it
was her own act, a strong revulsion of feeling caused the queen's
stately form to shake from head to foot: and in that supreme
moment she swore to her own turbulent heart that, come what
might, even to the fall of the Assyrian empire, Sarchedon should not
die!
She passed swiftly to the throne, and lifting the king's sceptre, laid
one end of it against her forehead, while she placed the other in his
hand.
"My lord," she said, "this is the feast of Baal. It is not lawful to slay
an Assyrian born during the worship of the great Assyrian god."
There shone a red light in the king's eyes that meant death, and the
foam stood on his lip. When he looked thus, it was in vain to sue for
pardon. Nevertheless, he passed his wrinkled hand over the fair
brow of the woman kneeling at his feet.
"Be it so," said Ninus. "To-morrow he shall die at sunrise. The king
hath spoken."
Then the guards looked furtively in each other's faces; for all men
knew from such a judgment there was no appeal, in such a sentence
no hope of mercy or reprieve.

CHAPTER XV
THE QUEEN'S PETITION
Sarchedon was hurried away in the custody of his former comrades,
who, pitying the fate their experience taught them was inevitable,
had yet discretion to take him from the presence of Ninus ere some
hideous cruelty or mutilation should be added to his punishment.
They were hardly out of the king's sight, however, when a priest of
Baal, arriving in breathless haste, brought an order from Assarac to
deliver up their prisoner in the temple of the god. On the festival of
that national deity, unusual respect was paid to the sacerdotal
character; and as, even amongst the guards of the Great King,
Assarac's policy had taught him to cultivate friendship and acquire
influence, the high priest's behest was obeyed readily, as if it had
emanated from Arbaces or even Ninus himself.
Sarchedon therefore became only so far a prisoner that he was not
permitted to pass the guards at any point of egress from the sacred
building, but might roam at large through its spacious chambers,
speculating on his chances of escape when night should fall, and he
could take advantage of such secret communications as his
knowledge of its votaries taught him must surely exist between the
temple and the town.
Meantime, however, he was a caged bird, yearning wildly for
freedom because of her whom he dearly loved. The queen's shaft
was shot deftly home, and the poison with which it had been tipped
did its work as cruelly as the pitiless archer could have desired. It
was madness to think of Ishtar in the arms of Ninyas; to feel that,
whilst he was a prisoner here, she might even be struggling for
personal freedom, perhaps calling on him to save her in vain.

But men trained to warfare acquire the habit of reviewing calmly all
sides of a dilemma, neither undervaluing its difficulties nor
despairing to vanquish them; especially they take into consideration
the bearing of probabilities and the important doctrine of chance. It
was not long before Sarchedon reflected he had himself seen
Arbaces under shield and helmet within a brief space of the queen's
arrival at her husband's palace; that if the espousals of his daughter
were really taking place with a prince, the chief captain would hardly
be absent from such a ceremony; and that Semiramis might have
thought it not below her dignity to tell him an absolute falsehood for
reasons of her own—reasons, he suspected, that ought to be
flattering to his self-love and conducive to the safety of his person. It
was impossible to mistake her avowed interest, her obvious
condescension, her changing moods and the bitterness with which
she accosted him in their late interview under the very eyes of the
Great King. If Semiramis loved him, he thought, she would surely
provide for his escape; and the first use he would make of his
freedom should be to seek Ishtar and urge her to fly with him at
once. Merodach could bear them both far beyond pursuit into the
desert, where they would find a hiding-place from the king's
merciless hatred and the queen's more cruel love.
Sarchedon, then, imprisoned in the temple of Baal, was hardly so ill
at ease as the wilful imperious woman whose reckless malice had
brought him to captivity and shame.
The old king scowled at her with fierce jealousy and rage as her
eyes followed the retiring form of the culprit, hurried out of the royal
presence with judicious promptitude by his comrades; but from the
first moment Ninus ever looked on that winsome face, he had found
in it a charm his heart was powerless to resist, and he was half
subdued already ere she leaned towards him with tender confiding
grace, and crossing her hands over his gaunt arm, rested her brow
on them, while she murmured in low soft accents,
"I thank my lord that he has turned no deaf ear to the voice of his
handmaiden. But enough of this. It is not well that Ninus should be

moved by the misconduct of a thoughtless spearman born under an
evil star. I have been summoned hastily to his presence. I feared he
was ill at ease. Is it overbold of his loving servant to ask what ails
my lord the king?"
"Nothing ails me," was the impatient answer; "nothing but the
clamour of women's voices and the senseless outcries of priests. I
sent for the queen," he added more gently, "because she is the light
of mine eyes and the priceless jewel of my treasure-house."
Semiramis rose erect, and bowing her lovely head, stood with her
hands crossed in the prescribed attitude of humility proper for a
subject.
She knew right well that in no position could she show to more
advantage; the pride of her bearing softened, the tender graces of
her womanhood enhanced, by its expression of shy compliance, of
loving submission to her lord.
"His servant hasted hither," said she, "on the instant the king's
command reached her palace. I had scarce time to tire my head and
smooth my robes. Yet I would fain look my best and proudest in the
sight of my lord the king."
He gazed on her with a fond admiration that was touching to see in
that war-worn old face, softening its rugged outlines and bringing
into the sunken eyes something of the wistful fidelity with which a
dog watches for the smile of its owner.
"Tired by a score of handmaidens," said he, "blazing in a hundred
jewels, or dishevelled and disrobed, with her free locks floating to
her knees, not the Queen of Heaven herself is to be compared to my
queen, fair and matchless to-day as on that bright morning when I
saw her ride through the camp like a vision, bow in hand, and
granted her the very first boon she asked me, for love of her sweet
face and her soft pleading eyes."

"And am I still so fair?" smiled the queen, while a flush of hope,
triumph, and pride in conscious beauty deepened the colour on her
cheek. "Nay, I shall scarce be brought to believe he is in earnest
unless I can prevail on my lord the king to grant me once again the
request I lay at his royal feet. If he loves me, surely he will not
refuse; and—and I think he loves me a little still!"
"I will have him flayed alive who gainsays it!" answered Ninus. "I
have ceased to love most things now, from the roar of battle to the
bubble of a wine-cup. But may I burn like a log of cedar in the fire of
Belus when I cease to love my queen!"
She shot at him one of those glances she could command at will, in
which mirth, tenderness, and modesty were blended with the fire of
love. "I believe it," she murmured gently. "Such an affection as ours
is written in the stars, and kindles into flame at the first meeting of
those who are destined for each other. It seems but yesterday that
my lord burst on my sight like Shamash, god of day, rising in
splendour on the camp, and I turned my head away to bury my
blushing face in my hands, because—because, already I loved him
only too well."
With the thrill that vibrated in every fibre of the old king's frame
arose the invariable accompaniment of sincere affection—a sense of
uncertainty and unworthiness.
"I was a stout warrior then," said he, "and not so uncomely, for one
whose life had been spent in saddle and war-chariot; but the colour
has faded on my cheek now, and worse, the fire has gone from my
spirit like the strength from my limbs."
There was a plaintive ring in the deep hoarse voice, that must have
touched any heart, save that of a woman with a purpose in view.
"Not so!" she exclaimed, hanging fondly about him. "Not so, my lord,
my love, my hero! I swear by the host of heaven, that to me you are
more noble, more kingly, more beautiful now, in the dignity of your
past deeds and mature fame, than in all the vehemence and ardour

of your impetuous manhood. Nay, my beloved," she added, half
playfully, half sadly, while clinging yet closer to his side, "it is not I
alone who think so; there were looks shot at my lord as he rode
through the streets from the brightest eyes in Babylon, that had I
not known full surely I was his only queen and love, would have
made me so miserable I had fled straightway to the desert, and
never looked on the face of man again."
Is there any age at which the male heart becomes insensible to such
flattery? With ebbing life and failing vigour, battered and out-worn
by a hundred battles, glorious in the splendour of a hundred
victories, the Great King might surely have been above that boyish
vanity, which counts for a triumph the empty gain of a woman's
fancy; yet Ninus smiled well pleased, and Semiramis felt that her
petition was already more than half granted, her game more than
half won.
"They know a stout spearman when they see one still," said the old
hero proudly, "and they judge by the ruin, doubtless, what the tower
must have been in its prime. Well, well, it stood many an assault in
its day, and from hosts of many nations, nor thought once of
surrender, till my queen here marched in and took possession, with
all the honours of war."
"And she has held it since against every woman in the world!"
murmured his wife, with another of those resistless glances, and a
bright flush. "Is it not so? Keep me not in the agony of suspense. Let
me have the king's word for my great happiness, and swear, by the
head of Nisroch, to grant me my desire!"
"I must hear first what it is," said the old warrior playfully; but
observing the tears start to her eyes, he added in fond haste, "Nay,
nay, beloved, the queen's petition shall be granted, whatever it be,
even to the half of mine empire."
"It is more than that!" exclaimed Semiramis, with a smile as ready as
her tears. "It is the whole empire I desire! I would fain sit in the seat

of my lord the king, but only for a day."
Ninus shook his head. "You are like your boy," said he fondly. "Do
you not remember when we took Ninyas for the first time to hunt
the lion outside the walls, and the lad must needs ride Samiel, the
wild war-horse, that bent to no hand but mine? By the blood of
Merodach, he wept like a maid, and I had not the heart to refuse
him; but when he was fairly in the saddle the tears soon dried on his
cheek, for the horse broke away with him like the wind of the desert,
from which he took his name. I tell you, while I stood there
dismounted, I must have felt what men call fear! I never knew how I
prized the boy, till my horse brought him back to me unhurt. Samiel
loved not to be far distant from his lord; and now Samiel is dead,
and his rider worn-out, and the queen—what was it the queen
asked? That she too should ride a steed she cannot control? Does
she know the pride of the Assyrian people, the turbulence of the
crowd, the daily clamour for sluices to be opened and granaries
unbarred, the craft of the priests, the false witness borne at the seat
of judgment, and the weight of the royal word, which may not be
recalled?"
But for the last consideration, the heart of Semiramis might have
been softened towards one who, with all his crimes and cruelties,
had yet been tender and loving in his home. The thought, however,
of Sarchedon's doom, ratified and rendered inevitable by those fatal
words, "The king hath spoken," swept all other considerations to the
winds, and she never looked truer, fairer, fonder than now, while she
answered in a tender whisper:
"My lord granted his request to our son at the sight of his wet eyes.
Shall he withhold from the mother her soul's desire, because she
cannot weep save when she fears to lose her place in the heart of
the Great King?"
His head sank on his breast; he was soon weary now, withering, as
it seemed, more hopelessly in the confinement of a palace than in

the freer atmosphere of a camp. "Name it," said he—"it is granted:
the king hath spoken."
Her eyes blazed with triumph, and the rich crimson mantled in her
cheek. "I have in my possession the signet of the Great King. I ask
to keep it until to-morrow at noon."
"I have said it," was the reply. "But what use will my queen make of
a toy that has often cumbered my hand more wearily than ever did
bridle, spear, or shield?"
"I will but use it to my lord's advantage," answered Semiramis
calmly. "Is not to-day the feast of Baal, and shall not the Great King
go up at nightfall into the cedar house on the roof to burn sacrifices,
and pour out drink-offerings before his god? There will be long
procession of priests, much leaping, howling, and gashing of
themselves at the altars; the prophets of the groves too must pass
before my lord, bearing earth and water, fir-cones, caskets, gold,
frankincense, and gifts. My lord is weary even now. Let him take his
rest undisturbed to strengthen him for the tedious labours of the
night. Meanwhile I hold the signet of the Great King and his
authority. I will provide for the safety of the nation, and for our
own."
He was getting drowsy, and his eyes were already half-closed.
"You have my signet," he murmured. "Send to Arbaces, and advise
with the chief captain for setting of the watch. And that
presumptuous spearman"—here he blazed up with an expiring flame
—"see that he be led forth at dawn. I have spoken, and he who
dared to cross the queen's path must die before the rise of another
day."
"Before the rise of another day!" she repeated mechanically; adding,
as she gathered her robes about her to depart, "I thank him that his
handmaiden hath found favour in his sight. I cover the feet of my
lord the king, and I take my leave."

But she turned at the great gate for one last look at the sleeping
form, mighty even in its ruin, and formidable in the abandonment of
its repose.
Proceeding from the palace, Semiramis paused to whisper a few
words in the ear of Arbaces. The chief captain seemed surprised,
and even discomposed by the purport of her communication; but
there was no appeal from a command backed by the royal signet,
and placing her hand, with the jewel in it, against his forehead, he
prostrated himself and withdrew. Had he remained, his discomfiture
might have been even greater to observe the queen in deep
consultation with Assarac, while Sargon, the king's shield-bearer,
remained, as if in waiting, a few paces off. The eunuch's head was
erect and his face bright with triumph; he wore the air of a man on
the eve of some great enterprise requiring skill, courage, and
intellect, but having at the same time perfect confidence in his own
power to carry it through.
"Is all ready?" asked Semiramis in a hollow whisper, while her cheek
paled, and a strange fire shone in her dark eyes.
"All is ready," answered the priest, in composed and measured
accents, as of one who states the details of a duty satisfactorily
fulfilled. "Double guards have been placed at the city gates; fifty
thousand archers, and as many spearmen, are mustered under
arms. Not a strained shaft nor a frayed bowstring amongst them,
and every man with his hand on his weapon, devoted to the queen's
interest for life and death!"
"We shall scarcely need them," was her reply. "I have commanded
Arbaces to remove his own especial power without the walls. Has my
son gone forth, and have you taken order for bestowing him in
safety to-night?"
"A company of spearmen will escort him," said the eunuch, "and will
guard the child and its new toy on the road to his refuge at Ascalon.
The king's signet will insure the obedience of such warriors as are

required to force the palace of Arbaces, and if the chief captain
resists with the strong hand, his blood be on his own head!"
"More slaughter!" exclaimed the queen sorrowfully. "O that the road
to power were not mired so deep with blood! But it is too late to
turn back now. Your life, my own, that poor condemned spearman of
the guard—all are at stake to-night; and we must not, we dare not,
stop. Is Sargon to be trusted? Yonder he stands, waiting for his
orders even now."
"Assarac glanced to where that warrior was stationed, a few paces
off, silent, erect, immovable, with the scowl of undying hatred on his
brow. The priest smiled—and the queen thought his smile more
fearful than the shield-bearer's frown—while he replied:
"A captive in the dungeon longs for light, and a gourd in the garden
for water; but what is their desire to a father's thirst for vengeance
on one who has shed the blood of his child?"

CHAPTER XVI
CRUEL AS THE GRAVE
The queen passed on a few paces without speaking, yet glanced
towards Assarac, who walked respectfully at her side, as though she
had something of importance on her mind. At last she observed
carelessly, "That spearman who has incurred the displeasure of my
lord the king. Is it not the messenger who brought me the royal
signet from the camp? These guards are all somewhat alike; yet I
seemed to recognise his face as he fell so untowardly at my feet."
"The same," answered Assarac, in his calm unmeaning tones. "A
goodly youth, and a stout warrior enough, by name Sarchedon. He
has been bestowed in the temple of Baal under my authority, safe at
least till nightfall. Nor can he escape, though guard and priest are
out of call; for there is no egress from the last chamber in the
painted gallery on the upper story where I have placed him, and
whence he could scarcely fly were he to borrow all the wings of
Nisroch, whose image stands over against the entrance to his
stronghold. But it is not of him I would speak," continued the priest,
keenly noting, though he never seemed to raise his eyes above the
hem of her garment, the queen's burning cheeks and air of
breathless interest. "From sunset to sunrise have I watched and
waited for the decree of the Seven Stars, poring over the scroll of
fire they unrolled for me, till my brain was giddy and mine eyes were
dim. Great Queen, there are no secrets in the future for him who has
learned to read the book of heaven. It teaches me that in the
darkness of this night shall dawn unclouded glory for the land of
Shinar, and supreme empire for her who is fairest and bravest
among women. As the goddess Ashtaroth is Queen of Heaven

above, so shall the great Semiramis be Queen of Earth below. The
Seven Stars have spoken it, and they cannot lie!"
He wondered at her preoccupation, contrasting with the attention
she had lately shown her present listlessness and apparent
indifference to the splendid destiny thus prophesied. Something
almost of scorn passed over his brow, while he reflected, that if the
mighty engine of ambition failed to move her intellect, he had yet a
subtler instrument with which to touch her heart.
Presently she roused herself to ask, "Did the stars promise only that
I should be great, or will they permit me also to be happy?"
"The queen's greatness," answered Assarac, "like her beauty, is
inseparable from her very being. Her happiness, like the robe that
covers it, can be put on or off at will."
"You are right," she exclaimed, while the resolute look he knew so
well passed over her beautiful face down to the very chin. "And she
who stands panting at a fountain were indeed a fool not to stoop
and drink. Tell me, then, their behests. What the stars bid me, that
will I do."
"The Great Queen cannot read from the book of heaven so readily
as a humble priest, the lowest of her slaves, though this lore, too,
will I aspire to teach her at some future time; but there lies in the
temple, fairly writ out in the Assyrian character and plain as the
flight of an arrow through the air, a scroll that teaches us poor
servants of Baal the rudiments of those mysteries into which the
ruler of a mighty empire must needs inquire. It is to be found in a
secure chamber of the painted gallery under the winged image of
Nisroch our god."
While he spoke, not the slightest curl of his lip, the faintest inflection
of his voice, betrayed a hidden motive, another meaning from that
which the plain straightforward words seemed to convey. Yet the
queen glanced very keenly in his face, while she stopped short in her
walk and turned towards the temple, observing only—

"It is not yet near sunset. I shall have light to read the scroll."
Then she dismissed Kalmim and her women, desiring that she might
be attended only by the priest of Baal, in whose steps, nevertheless,
Sargon followed like his shadow.
Arrived within the porch of the temple, she gave a great sigh of
relief, as though she luxuriated in the refreshing coolness of those
spacious halls, with their smooth shining floors, their countless
columns, their vast shadowy recesses, that spoke of calm and
secrecy and repose. She had not gone far, ere Assarac stopped and
prostrated himself at her feet.
"Let not the queen be wroth with the lowest of her servants," said
the wily eunuch, "if he ask permission to be relieved for a brief
space from attendance on her person. There is so much to be
prepared for the feast of Baal, so many details to arrange for the
sacrifice of to-night, that I must neglect my duties no longer. The
scroll lies where all who pass may read, and when the Great Queen
has studied it enough, if, standing in this spot, she will but clap her
hands thus, those shall be within call who can summon me to her
presence without delay."
Semiramis frowned, though the frown did but mask a smile.
"It is scarce a royal reception," said she; "nevertheless, be it so. I
am content to breathe this cool and grateful air for a space, ere I
return with Kalmim and the women to my palace across the river.
You are dismissed."
He rose and retired, making a sign to Sargon, who watched his
every movement, that caused the shield-bearer to follow him
forthwith.
Clear of the queen's presence, Assarac pointed to a table on which
stood a golden flagon and drinking-cups of the same metal.
"Not even to-day?" said he, while the other shook his head in token
of dissent. "Trust me, Sargon, you will be faint and athirst before all

is done."
"Not a drop of wine shall cross my lips," answered the shield-bearer
in a fierce determined whisper, "till I have dipped my hands in the
blood of him who has injured me. I have sworn it by the splendour
of Nisroch. It is the oath of the Great King!"
"Is your vengeance, then, so deadly?" asked the eunuch, in a tone
of pity that obviously chafed and aggravated the passion it seemed
to commiserate. "Surely ten score of sheep, five yoke of oxen, a
hundred camel-loads of barley, or a talent of gold should absolve the
shedder of blood from farther reparation. In our land of Shinar the
laws are merciful, and do not exact life for life."
"There is a law in man's heart," replied Sargon, still in the same low
concentrated accents, "that sets aside the law of nations and the
artificial ordinances of priests. See here," he continued, plucking
from his girdle a knotted bowstring, limp and frayed, which he put in
the other's hand; "a reader of the stars should be able to tell a
simple spearman how many knots on that bit of twisted silk go to
the score."
"It needs no great study to perceive that but one is left here now,"
answered Assarac with an inquiring look into the other's face.
"The bow from which I took that string had been bent many a time
in the Great King's service," was the reply; "and a shaft it sped but
seldom missed its mark. I have covered Ninus under shield, and
defended him with my body, when arrows and javelins were flying
thick as the sands of the desert before a south wind. I have waged
my life, poured out my blood freely for my lord, and he has
rewarded me with his own royal hand."
"He is lavish enough," observed Assarac, "be it gold or stripes,
honours or death, that he awards. May the king live for ever!"
"May the king live for ever!" repeated his shield-bearer, "a god
among gods, a star in the host of heaven. If an empty throne be

waiting for him up yonder, may it soon be filled! When I saw my boy
fall stark dead, the blood gushing from his mouth and nostrils, I
prostrated myself and did obeisance to the Great King; but I drew
that string from my bow, and in it I tied a score of knots, swearing
with each a deadly oath, that by the splendour of Nisroch I would be
avenged ere the twentieth was undone. Since then I have loosed a
knot with every sunrise; and lo, a priest of Baal counts, and tells me
there is but one left!"
Beneath its sallow skin a terrible smile rounded the fleshy outlines of
the eunuch's face. His voice, however, remained firm while he
whispered—
"We understand each other, and there must be no wavering—no
escape—no mercy!"
Between his clenched teeth the shield-bearer's answer came in
single syllables, hissing like drops of blood on a burning hearth—
"Such wavering as stayed the cruel hand, the deadly bow! Such
escape as was afforded that light-footed youth, whom only an
arrow's flight could overtake! Such mercy as he showed my boy!"
"Come with me," was the high-priest's reply; and the two ascended
a spiral staircase of carved and polished wood-work, leading to the
Talar or cedar-chamber on the roof of the temple, where at nightfall
sacrifice was to be offered, and drink-offerings poured out in person
by the Great King to his Assyrian god. Here they drew from a store-
chamber within the wall several bundles of reeds, which they
strewed in profusion over the wooden floor of the cedar-house, and
which Assarac sprinkled assiduously with a certain fluid from a phial
he had kept hidden beneath his gown.
"Every precaution must be taken," observed the priest with another
hideous smile. "But if it be the will of his ancestor Ashur to descend
for him in a chariot of fire, and these reeds thus saturated should
catch the flame, then must the Great King, if he be not overcome

Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.
More than just a book-buying platform, we strive to be a bridge
connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.
Join us on a journey of knowledge exploration, passion nurturing, and
personal growth every day!
ebookbell.com

Intelligent Computing Proceedings Of The 2020 Computing Conference Volume 3 1st Ed Kohei Arai

About This Presentation

Slide Content

Tags

Categories

Download

Quick Actions

Statistics

Related Slideshows

Intelligent Computing Proceedings Of The 2020 Computing Conference Volume 3 1st Ed Kohei Arai

About This Presentation

Slide Content

Slide 1

Slide 2

Slide 3

Slide 5

Slide 6

Slide 7

Slide 8

Slide 9

Slide 10

Slide 11

Slide 12

Slide 13

Slide 14

Slide 15

Slide 16

Slide 17

Slide 18

Slide 19

Slide 20

Slide 21

Slide 22

Slide 23

Slide 24

Slide 25

Slide 26

Slide 27

Slide 28

Slide 29

Slide 30

Slide 31

Slide 32

Slide 33

Slide 34

Slide 35

Slide 36

Slide 37

Slide 38

Slide 39

Slide 40

Slide 41

Slide 42

Slide 43

Slide 44

Slide 45

Slide 46

Slide 47

Slide 48

Slide 49

Slide 50

Slide 51

Slide 52

Slide 53

Slide 54

Slide 55

Slide 56

Slide 57

Slide 58

Slide 59

Slide 60

Slide 61

Slide 62

Slide 64

Slide 65

Slide 66

Slide 67

Slide 68

Slide 69

Slide 70

Slide 71

Slide 72

Slide 73

Slide 74

Slide 75

Slide 76

Slide 77

Slide 78

Tags