Artificial Intelligence Applications and

Innovations: 15th IFIP WG 12.5

International Conference, AIAI 2019,
Hersonissos, Crete, Greece, May 24–26,
2019, Proceedings John Macintyre
Visit to download the full and correct content document:
https://textbookfull.com/product/artificial-intelligence-applications-and-innovations-15t
h-ifip-wg-12-5-international-conference-aiai-2019-hersonissos-crete-greece-may-24-2
6-2019-proceedings-john-macintyre/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Artificial Intelligence Applications and Innovations:

15th IFIP WG 12.5 International Conference, AIAI 2019,
Hersonissos, Crete, Greece, May 24–26, 2019,
Proceedings John Macintyre
https://textbookfull.com/product/artificial-intelligence-
applications-and-innovations-15th-ifip-wg-12-5-international-
conference-aiai-2019-hersonissos-crete-greece-
may-24-26-2019-proceedings-john-macintyre/

Engineering Applications of Neural Networks: 20th

International Conference, EANN 2019, Xersonisos, Crete,
Greece, May 24-26, 2019, Proceedings John Macintyre

https://textbookfull.com/product/engineering-applications-of-
neural-networks-20th-international-conference-
eann-2019-xersonisos-crete-greece-may-24-26-2019-proceedings-
john-macintyre/

Artificial Intelligence Applications and Innovations

16th IFIP WG 12 5 International Conference AIAI 2020
Neos Marmaras Greece June 5 7 2020 Proceedings Part I
Ilias Maglogiannis
https://textbookfull.com/product/artificial-intelligence-
applications-and-innovations-16th-ifip-wg-12-5-international-
conference-aiai-2020-neos-marmaras-greece-
june-5-7-2020-proceedings-part-i-ilias-maglogiannis/

Cloud Computing and Services Science 9th International

Conference CLOSER 2019 Heraklion Crete Greece May 2 4
2019 Revised Selected Papers Donald Ferguson

https://textbookfull.com/product/cloud-computing-and-services-
science-9th-international-conference-closer-2019-heraklion-crete-
greece-may-2-4-2019-revised-selected-papers-donald-ferguson/
Artificial Intelligence for Communications and
Networks: First EAI International Conference, AICON
2019, Harbin, China, May 25–26, 2019, Proceedings, Part
I Shuai Han
https://textbookfull.com/product/artificial-intelligence-for-
communications-and-networks-first-eai-international-conference-
aicon-2019-harbin-china-may-25-26-2019-proceedings-part-i-shuai-
han/

Artificial Intelligence for Communications and

Networks: First EAI International Conference, AICON
2019, Harbin, China, May 25–26, 2019, Proceedings, Part
II Shuai Han
https://textbookfull.com/product/artificial-intelligence-for-
communications-and-networks-first-eai-international-conference-
aicon-2019-harbin-china-may-25-26-2019-proceedings-part-ii-shuai-
han/

Artificial Intelligence XXXVI 39th SGAI International

Conference on Artificial Intelligence AI 2019 Cambridge
UK December 17 19 2019 Proceedings Max Bramer

https://textbookfull.com/product/artificial-intelligence-
xxxvi-39th-sgai-international-conference-on-artificial-
intelligence-ai-2019-cambridge-uk-
december-17-19-2019-proceedings-max-bramer/

Testing Software and Systems 31st IFIP WG 6 1

International Conference ICTSS 2019 Paris France
October 15 17 2019 Proceedings Christophe Gaston

https://textbookfull.com/product/testing-software-and-
systems-31st-ifip-wg-6-1-international-conference-
ictss-2019-paris-france-october-15-17-2019-proceedings-
christophe-gaston/

Trust Management XIII 13th IFIP WG 11 11 International

Conference IFIPTM 2019 Copenhagen Denmark July 17 19
2019 Proceedings Weizhi Meng

https://textbookfull.com/product/trust-management-xiii-13th-ifip-
wg-11-11-international-conference-ifiptm-2019-copenhagen-denmark-
july-17-19-2019-proceedings-weizhi-meng/
 IFIP AICT 559
John MacIntyre
Ilias Maglogiannis
Lazaros Iliadis
Elias Pimenidis
(Eds.)

Artificial Intelligence
Applications
and Innovations

15th IFIP WG 12.5 International Conference, AIAI 2019

Hersonissos, Crete, Greece, May 24–26, 2019
Proceedings

123
IFIP Advances in Information
and Communication Technology 559

Editor-in-Chief

Kai Rannenberg, Goethe University Frankfurt, Germany

Editorial Board Members

TC 1 – Foundations of Computer Science
Jacques Sakarovitch, Télécom ParisTech, France
TC 2 – Software: Theory and Practice
Michael Goedicke, University of Duisburg-Essen, Germany
TC 3 – Education
Arthur Tatnall, Victoria University, Melbourne, Australia
TC 5 – Information Technology Applications
Erich J. Neuhold, University of Vienna, Austria
TC 6 – Communication Systems
Aiko Pras, University of Twente, Enschede, The Netherlands
TC 7 – System Modeling and Optimization
Fredi Tröltzsch, TU Berlin, Germany
TC 8 – Information Systems
Jan Pries-Heje, Roskilde University, Denmark
TC 9 – ICT and Society
David Kreps, University of Salford, Greater Manchester, UK
TC 10 – Computer Systems Technology
Ricardo Reis, Federal University of Rio Grande do Sul, Porto Alegre, Brazil
TC 11 – Security and Privacy Protection in Information Processing Systems
Steven Furnell, Plymouth University, UK
TC 12 – Artificial Intelligence
Ulrich Furbach, University of Koblenz-Landau, Germany
TC 13 – Human-Computer Interaction
Marco Winckler, University of Nice Sophia Antipolis, France
TC 14 – Entertainment Computing
Rainer Malaka, University of Bremen, Germany
IFIP – The International Federation for Information Processing
IFIP was founded in 1960 under the auspices of UNESCO, following the first World
Computer Congress held in Paris the previous year. A federation for societies working
in information processing, IFIP’s aim is two-fold: to support information processing in
the countries of its members and to encourage technology transfer to developing na-
tions. As its mission statement clearly states:

IFIP is the global non-profit federation of societies of ICT professionals that aims
at achieving a worldwide professional and socially responsible development and
application of information and communication technologies.

IFIP is a non-profit-making organization, run almost solely by 2500 volunteers. It

operates through a number of technical committees and working groups, which organize
events and publications. IFIP’s events range from large international open conferences
to working conferences and local seminars.
The flagship event is the IFIP World Computer Congress, at which both invited and
contributed papers are presented. Contributed papers are rigorously refereed and the
rejection rate is high.
As with the Congress, participation in the open conferences is open to all and papers
may be invited or submitted. Again, submitted papers are stringently refereed.
The working conferences are structured differently. They are usually run by a work-
ing group and attendance is generally smaller and occasionally by invitation only. Their
purpose is to create an atmosphere conducive to innovation and development. Referee-
ing is also rigorous and papers are subjected to extensive group discussion.
Publications arising from IFIP events vary. The papers presented at the IFIP World
Computer Congress and at open conferences are published as conference proceedings,
while the results of the working conferences are often published as collections of se-
lected and edited papers.
IFIP distinguishes three types of institutional membership: Country Representative
Members, Members at Large, and Associate Members. The type of organization that
can apply for membership is a wide variety and includes national or international so-
cieties of individual computer scientists/ICT professionals, associations or federations
of such societies, government institutions/government related organizations, national or
international research institutes or consortia, universities, academies of sciences, com-
panies, national or international associations or federations of companies.

More information about this series at http://www.springer.com/series/6102

John MacIntyre Ilias Maglogiannis
• •

Lazaros Iliadis Elias Pimenidis (Eds.)

•

Artificial Intelligence
Applications
and Innovations
15th IFIP WG 12.5 International Conference, AIAI 2019
Hersonissos, Crete, Greece, May 24–26, 2019
Proceedings

123
Editors
John MacIntyre Ilias Maglogiannis
University of Sunderland University of Piraeus
Sunderland, UK Piraeus, Greece
Lazaros Iliadis Elias Pimenidis
Democritus University of Thrace University of West England
Xanthi, Greece Bristol, UK

ISSN 1868-4238 ISSN 1868-422X (electronic)

IFIP Advances in Information and Communication Technology
ISBN 978-3-030-19822-0 ISBN 978-3-030-19823-7 (eBook)
https://doi.org/10.1007/978-3-030-19823-7

© IFIP International Federation for Information Processing 2019

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
 Preface

AIAI 2019

According to Professor Klaus Schwab (founder and executive chairman of the World
Economic Forum), we are living in the era of a great revolution that is rapidly bringing
huge changes and challenges in our daily lives. This is the Fourth Industrial Revolu-
tion, which has a big impact on all disciplines, even in the way that we communicate
and interact with each other. Artificial intelligence (AI) is a major and significant part
of the Fourth Industrial Revolution. Its rapid technical breakthroughs are enabling
superhuman performance by machines in a real-time mode. Machine vision (e.g., face
recognition) or language translators and assistants like Siri and Alexa are characteristic
examples. AI is promising a brave new world where business and economies will
expand their productivity and innovation. Machine learning and deep learning are part
of our usual common interactions on our mobile phones and on social media.
Numerous applications of AI are used in almost all domains from cybersecurity to
financial and medical cases. However, historic challenges for the future of mankind are
being faced. Potential unethical use of AI may violate democratic human rights and
may alter the character of Western societies.
The 15th Artificial Intelligence Applications and Innovations (AIAI) conference
offered insight into all timely challenges related to technical, legal, and ethical aspects
of intelligent systems and their applications. New algorithms and potential prototypes
employed in diverse domains were introduced.
AIAI is a mature international scientific conference held in Europe and is well
established in the scientific area of AI. Its history is long and very successful, following
and spreading the evolution of intelligent systems.
The first event was organized in Toulouse France in 2004. Since then, it has had a
continuous and dynamic presence as a major global, but mainly European scientific
event. More specifically, it has been organized in China, Greece, Cyprus, Australia, and
France. It has always been technically supported by the International Federation for
Information Processing (IFIP) and more specifically by the Working Group 12.5, which
is interested in AI applications.
Following a long-standing tradition, this Springer volume belongs to the IFIP AICT
Springer Series and it contains the papers that were accepted to be presented orally at
the AIAI 2019 conference. An additional volume comprises the papers that were
accepted and presented at the workshops and were held as parallel events.
vi Preface

The diverse nature of papers presented demonstrates the vitality of AI algorithms

and approaches. It certainly proves the very wide range of AI applications as well.
The event was held during May 24–26, 2019, in the Aldemar Knossos Royal
five-star Hotel in Crete, Greece.
The response of the international scientific community to the AIAI 2019 call for
papers was more than satisfactory, with 101 papers initially submitted. All papers were
peer reviewed by at least two independent academic referees. Where needed, a third
referee was consulted to resolve any potential conflicts. A total of 49 papers (48.5%
of the submitted manuscripts) were accepted to be published as full papers (12 pages
long) in the proceedings. Owing to the high quality of the submissions, the Program
Committee decided that it should accept six more papers to be published as short ones
(10 pages long).
Three scientific workshops on timely AI subjects were organized under the
framework of AIAI 2019.
– The 8th Mining Humanistic Data Workshop (MHDW 2019) organized by the
University of Patras and Ionion University, Greece
– The 4th Workshop on 5G-Putting Intelligence to the Network Edge (5G-PINE
2019) organized by the research team of the Hellenic Telecommunications Orga-
nization (OTE) in cooperation with 22 major partner companies
– The First Workshop on Emerging Trends in AI (ETAI 2019)
Sponsored by the Springer journal Neural Computing and Applications
(open workshop without submission of papers)
We are grateful to Professor John MacIntyre from the University of Sunderland,
UK, for organizing this workshop and, moreover, for his continuous support of the
AIAI and EANN conferences.
We wish to thank Professor Andrew Starr for his contribution to this very interesting
workshop.
AI is in a new “boom” period, with exponential growth in commercialization of
research and development, products being introduced into the market with embedded
AI as well as “intelligent systems” of various types. Projections for commercial
revenue from AI show exponential growth; such is the ubiquitous nature of AI in the
modern world that members of the public are interacting with intelligent systems or
agents every day – even though they often are not aware of it!
This workshop, led by Professor John MacIntyre, considered emerging themes in
AI, covering not only the technical aspects of where AI is going, but the wider question
of ethics, and the potential for future regulatory frameworks for the development,
implementation, and operation of intelligent systems and their role in our society.
The workshop format included three short presentations by the keynote speakers,
followed by an interactive panel Q&A session where the panel members and audience
engaged in a lively debate on the topics discussed!
The subjects of their presentations were the following:
– John MacIntyre: “The Future of AI – Existential Threat or New Revolution?”
– Andrew Starr: “Practical AI for Practical Problems”
 Preface vii

This was an open workshop without submission of papers.

Four keynote speakers were invited to give lectures on timely aspects of AI. The
following talks were given:
1. Professor Plamen Angelov, University of Lancaster, UK: “Empirical Approach:
How to Get Fast, Interpretable Deep Learning”
2. Dr. Evangelos Eleftheriou, IBM Fellow, Cloud and Computing Infrastructure,
Zurich Research Laboratory Switzerland. subject: “In-Memory Computing:
Accelerating AI Applications”
3. Dr. John Oommen: Carleton University, Ottawa, Canada: “The Power of Pursuit.
Learning Paradigm in the Partitioning of Data”
4. Professor Panagiotis Papapetrou, Stockholm University: “Learning from Electronic
Health Records: From temporal Abstraction to Timeseries Interpretability”
A tutorial was hosted on the topic: “Automated Machine Learning for Bioinfor-
matics and Computational Biology.”
The tutorial (3 hours) was given by Professor Ioannis Tsamdinos (Computer Science
Department of University of Crete, co-founder of Gnosis Data Analysis PC, a
University spin-off company, and Affiliated Faculty at IACM-FORTH) and Professor
Vincenzo Lagani (Ilia State University, Tbilisi, Georgia and Gnosis Data Analysis PC
co-founder).
Numerous bioinformaticians, computational biologists, and life scientists in general
are applying supervised learning techniques and feature selection in their research
work. The tutorial was addressed to this audience intending to shield them against
methodological pitfalls, inform them about new methodologies and tools emerging in
the field of Auto-ML, and increase their productivity.
The accepted papers of the 15th AIAI conference are related to the following
thematic topics:
Deep learning ANN
Genetic algorithms - optimization
Constraints modeling
ANN training algorithms
Social media intelligent modeling
Text mining/machine translation
Fuzzy modeling
Biomedical and bioinformatics algorithms and systems
Feature selection
Emotion recognition
Hybrid intelligent models
Classification-pattern recognition
Intelligent security modeling
Complex stochastic games
Unsupervised machine learning
ANN in industry
Intelligent clustering
Convolutional and recurrent ANN
viii Preface

Recommender systems
Intelligent telecommunications modeling
Intelligent hybrid systems using internet of things
The authors of submitted papers came from 23 different countries from all over the
globe, namely: Austria, Brazil, Canada, Colombia, Czech Republic, Finland, France,
Germany, Greece, The Netherlands, Hungary, India, Italy, Japan, P.R. China, Peru,
Poland, Portugal, Spain, Sweden, Tunisia, the UK, and the USA.

May 2019 John MacIntyre

Ilias Maglogiannis
Lazaros Iliadis
Elias Pimenidis
 Organization

Executive Committee
General Chairs
John MacIntyre University of Sunderland, UK (Dean of the Faculty
of Applied Sciences and Pro Vice Chancellor
of the University of Sunderland)
Ilias Maglogiannis University of Piraeus, Greece
(President of the
IFIP WG12.5)
Plamen Angelov University of Lancaster, UK

Program Chairs
Lazaros Iliadis Democritus University of Thrace, Greece
Elias Pimenidis University of the West of England, Bristol, UK

Advisory Chairs
Stefanos Kolias University of Lincoln, UK
Spyros Likothanasis University of Patras, Greece
Georgios Vouros University of Piraeus, Greece

Honorary Chair
Barbara Hammer Bielefeld University, Germany

Workshop Chairs
Christos Makris University of Patras, Greece
Phivos Mylonas Ionian University, Greece
Spyros Sioutas University of Patras, Greece

Publication and Publicity Chair

Antonis Papaleonidas Democritus University of Thrace, Greece

Program Committee
Michel Aldanondo IMT Mines Albi, France
Athanasios Alexiou NGCEF, Australia
Mohammed Alghwell Freelancer, Libya
Ioannis Anagnostopoulos University of Central Greece, Greece
George Anastassopoulos Democritus University of Thrace, Greece
Vardis-Dimitris Anezakis Democritus University of Thrace, Greece
x Organization

Costin Badica University of Craiova, Romania

Kostas Berberidis University of Patras, Greece
Nik Bessis Edge Hill University, UK
Varun Bhatt Indian Institute of Technology, Bombay, India
Giacomo Boracchi Politecnico di Milano, Italy
Farah Bouakrif University of Jijel, Algeria
Antonio Braga Federal University of Minas Gerais, Brazil
Peter Brida University of Zilina, Slovakia
Ivo Bukovsky Tohoku University, Japan
Paulo Vitor Campos Souza CEFET-MG, Brazil
George Caridakis National Technical University of Athens, Greece
Jheymesson Cavalcanti UPE, Brazil
Ioannis Chamodrakas National and Kapodistrian University of Athens,
Greece
Ioannis Chochliouros Hellenic Telecommunications Organization S.A.
(OTE), Greece
Adriana Mihaela Coroiu Babes Bolyai University
Dawei Dai Fudan University, China
Vilson Luiz Dalle Mole UTFPR, Brazil
Debasmit Das Purdue University, USA
Bodhisattva Dash IIIT Bhubaneswar, India
Konstantinos Demertzis Democritus University of Thrace, Greece
Antreas Dionysiou University of Cyprus, Cyprus
Ioannis Dokas DUTH, Greece
Sergey Dolenko D.V. Skobeltsyn Institute of Nuclear Physics,
M.V. Lomonosov Moscow State University, Russia
Xiao Dong Institute of Computing Technology, China
Shirin Dora University Van Amsterdam, The Netherlands
Rodrigo Exterkoetter LTrace Geophysical Solutions
Mauro Gaggero National Research Council, Italy
Claudio Gallicchio University of Pisa, Italy
Ignazio Gallo University of Insubria, Italy
Spiros Georgakopoulos University of Thessaly, Greece
Eleonora Giunchiglia Università di Genova, Italy
Giorgio Gnecco IMT School for Advanced Studies, Italy
Ioannis Gkourtzounis University of Northampton, Greece
Foteini Grivokostopoulou University of Patras, Greece
Hakan Haberdar University of Houston, USA
Petr Hajek University of Pardubice, Czech Republic
Xue Han China University of Geosciences, China
Ioannis Hatzilygeroudis University of Patras, Greece
Jian Hou Bohai University, China
Lazaros Iliadis Democritus University of Thrace, Greece
Jacek Kabziński Lodz University of Technology, Poland
Antonios Kalampakas AUM, Kuwait
Andreas Kanavos University of Patras, Greece
 Organization xi

Savvas Karatsiolis University of Cyprus, Cyprus

Kostas Karatzas Aristotle University of Thessaloniki, Greece
Antonios Karatzoglou Karlsruhe Institute of Technology, Germany
Ioannis Karydis Ionian University, Greece
Petros Kefalas University of Sheffield International Faculty, Greece
Katia Lida Kermanidis Ionian University, Greece
Nadia Masood Khan University of Engineering and Technology Peshawar,
Pakistan
Sophie Klecker University of Luxembourg, Luxembourg
Yiannis Kokkinos University of Macedonia, Greece
Petia Koprinkova-Hristova Bulgarian Academy of Sciences, Bulgaria
Athanasios Koutras TEI of Western Greece, Greece
Ondrej Krejcar University of Hradec Kralove, Czech Republic
Efthyvoulos Kyriacou Frederick University, Cyprus
Guangli Li Institute of Computing Technology, Chinese Academy
of Sciences, China
Annika Lindh Dublin Institute of Technology, Ireland
Ilias Maglogiannis University of Piraeus, Greece
George Magoulas University of London, Birkbeck College, UK
Christos Makris University of Patras, Greece
Mario Malcangi Università degli Studi di Milano, Italy
Boudjelal Meftah University Mustapha Stambouli, Mascara, Algeria
Nikolaos Mitianoudis Democritus University of Thrace, Greece
Haralambos Mouratidis University of Brighton, UK
Phivos Mylonas National Technical University of Athens, Greece
Shigang Yue University of Lincoln, UK
Yancho Todorov Aalto University, Espoo, Finland
George Tsekouras University of the Aegean, Greece
Mihaela Oprea Petroleum-Gas University of Ploiesti, Romania
Paul Krause University of Surrey, UK
Rafet Sifa Fraunhofer IAIS, Germany
Alexander Ryjov Lomonosov Moscow State University, Russia
Giannis Nikolentzos Ecole Polytechnique, France
Duc-Hong Pham VNU, Vietnam
Elias Pimenidis University of the West of England, UK
Hongyu Li Zhongan Tech, China
Marcello Sanguineti University of Genoa, Italy
Zhongnan Zhang Xiamen University, China
Doina Logofatu Frankfurt University of Applied Sciences, Germany
Ruggero Labati Università degli Studi di Milano, Italy
Florin Leon Technical University of Iasi, Romania
Aristidis Likas University of Ioannina, Greece
Spiros Likothanassis University of Patras, Greece
Francesco Marcelloni University of Pisa, Italy
Giorgio Morales INICTEL-UNI, Peru
Stavros Ntalampiras University of Milan, Italy
xii Organization

Basil Papadopoulos Democritus University of Thrace, Greece

Antonios Papaleonidas DUTH, Greece
Isidoros Perikos University of Patras, Greece
Nicolai Petkov University of Groningen, The Netherlands
Miltos Petridis Middlesex University, UK
Jielin Qiu Shanghai Jiao Tong University, China
Juan Qiu Tongji University, China
Bernardete Ribeiro University of Coimbra, Portugal
Simone Scardapane Sapienza University of Rome, Italy
Andreas Stafylopatis National Technical University of Athens, Greece
Antonino Staiano Parthenope University of Naples, Italy
Ioannis Stephanakis Hellenic Telecommunications Organisation SA, Greece
Ricardo Tanscheit PUC-Rio, Brazil
Francesco Trovò Politecnico di Milano, Italy
Nicolas Tsapatsoulis Cyprus University of Technology, Cyprus
Nikolaos Vassilas TEI of Athens, Greece
Petra Vidnerová The Czech Academy of Sciences, Czech Republic
Panagiotis Vlamos Ionian University, Greece
George Vouros University of Piraeus, Greece
Xin-She Yang Middlesex University, UK
Drago Žagar University of Osijek, Croatia
Rabiaa Zitouni University of Tunis el Manar, Tunisia
Abstracts of Invited Talks
 Learning from Electronic Health Records:
From Temporal Abstractions to Time Series
Interpretability

Panagiotis Papapetrou

Department of Computer and Systems Sciences, Stockholm University

panagiotis@dsv.su.se

Abstract. The first part of the talk will focus on data mining methods for
learning from Electronic Health Records (EHRs), which are typically perceived
as big and complex patient data sources. On them, scientists strive to perform
predictions on patients’ progress, to understand and predict response to therapy,
to detect adverse drug effects, and many other learning tasks. Medical
researchers are also interested in learning from cohorts of population-based
studies and of experiments. Learning tasks include the identification of disease
predictors that can lead to new diagnostic tests and the acquisition of insights on
interventions. The talk will elaborate on data sources, methods, and case studies
in medical mining.
The second part of the talk will tackle the issue of interpretability and
explainability of opaque machine learning models, with focus on time series
classification. Time series classification has received great attention over the past
decade with a wide range of methods focusing on predictive performance by
exploiting various types of temporal features. Nonetheless, little emphasis has
been placed on interpretability and explainability. This talk will formulate the
novel problem of explainable time series tweaking, where, given a time series
and an opaque classifier that provides a particular classification decision for the
time series, the objective is to find the minimum number of changes to be
performed to the given time series so that the classifier changes its decision to
another class. Moreover, it will be shown that the problem is NP-hard. Two
instantiations of the problem will be presented. The classifier under investigation
will be the random shapelet forest classifier. Moreover, two algorithmic solu-
tions for the two problem instantiations will be presented along with simple
optimizations, as well as a baseline solution using the nearest neighbor classifier.
 Empirical Approach: How to Get Fast,
Interpretable Deep Learning

Plamen Angelov

Department of Computing and Communications, University of Lancaster

p.angelov@lancaster.ac.uk

Abstract. We are witnessing an explosion of data (streams) being generated and

growing exponentially. Nowadays we carry in our pockets Gigabytes of data in
the form of USB flash memory sticks, smartphones, smartwatches etc.
Extracting useful information and knowledge from these big data streams is of
immense importance for the society, economy and science. Deep Learning
quickly become a synonymous of a powerful method to enable items and
processes with elements of AI in the sense that it makes possible human like
performance in recognizing images and speech. However, the currently used
methods for deep learning which are based on neural networks (recurrent, belief,
etc.) is opaque (not transparent), requires huge amount of training data and
computing power (hours of training using GPUs), is offline and its online ver-
sions based on reinforcement learning has no proven convergence, does not
guarantee same result for the same input (lacks repeatability).
The speaker recently introduced a new concept of empirical approach to
machine learning and fuzzy sets and systems, had proven convergence for a
class of such models and used the link between neural networks and fuzzy
systems (neuro-fuzzy systems are known to have a duality from the radial basis
functions (RBF) networks and fuzzy rule based models and having the key
property of universal approximation proven for both).
In this talk he will present in a systematic way the basics of the newly
introduced Empirical Approach to Machine Learning, Fuzzy Sets and Systems
and its applications to problems like anomaly detection, clustering, classifica-
tion, prediction and control. The major advantages of this new paradigm are the
liberation from the restrictive and often unrealistic assumptions and require-
ments concerning the nature of the data (random, deterministic, fuzzy), the need
to formulate and assume a priori the type of distribution models, membership
functions, the independence of the individual data observations, their large
(theoretically infinite) number, etc.
From a pragmatic point of view, this direct approach from data (streams) to
complex, layered model representation is automated fully and leads to very
efficient model structures. In addition, the proposed new concept learns in a way
similar to the way people learn – it can start from a single example. The reason
why the proposed new approach makes this possible is because it is prototype
based and non-parametric.
 “In-memory Computing”: Accelerating AI
Applications

Evangelos Eleftheriou

IBM Fellow, Cloud and Computing Infrastructure, Zurich Research Laboratory,

Zurich, Switzerland
ele@zurich.ibm.com

Abstract. In today’s computing systems based on the conventional von

Neumann architecture, there are distinct memory and processing units.
Performing computations results in a significant amount of data being moved
back and forth between the physically separated memory and processing units.
This costs time and energy, and constitutes an inherent performance bottleneck.
It is becoming increasingly clear that for application areas such as AI (and
indeed cognitive computing in general), we need to transition to computing
architectures in which memory and logic coexist in some form. Brain-inspired
neuromorphic computing and the fascinating new area of in-memory computing
or computational memory are two key non-von Neumann approaches being
researched. A critical requirement in these novel computing paradigms is a
very-high-density, low-power, variable-state, programmable and non-volatile
nanoscale memory device. There are many examples of such nanoscale memory
devices in which the information is stored either as charge or as resistance.
However, one particular example is phase-change-memory (PCM) devices,
which are very well suited to address this need, owing to their multi-level
storage capability and potential scalability.
In in-memory computing, the physics of the nanoscale memory devices, as
well as the organization of such devices in cross-bar arrays, are exploited to
perform certain computational tasks within the memory unit. I will present how
computational memories accelerate AI applications and will show small- and
large-scale experimental demonstrations that perform high-level computational
primitives, such as ultra-low-power inference engines, optimization solvers
including compressed sensing and sparse coding, linear solvers and temporal
correlation detection. Moreover, I will discuss the efficacy of this approach to
efficiently address not only inferencing but also training of deep neural net-
works. The results show that this co-existence of computation and storage at the
nanometer scale could be the enabler for new, ultra-dense, low-power, and
massively parallel computing systems. Thus, by augmenting conventional
computing systems, in-memory computing could help achieve orders of mag-
nitude improvement in performance and efficiency.
 Contents

Invited Paper

The Power of the “Pursuit” Learning Paradigm in the Partitioning of Data . . . 3

Abdolreza Shirvani and B. John Oommen

AI Anomaly Detection - Active Learning

Cyber-Typhon: An Online Multi-task Anomaly Detection Framework . . . . . . 19

Konstantinos Demertzis, Lazaros Iliadis, Panayiotis Kikiras,
and Nikos Tziritas

Investigating the Benefits of Exploiting Incremental Learners Under Active

Learning Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Stamatis Karlos, Vasileios G. Kanas, Nikos Fazakis, Christos Aridas,
and Sotiris Kotsiantis

The Blockchain Random Neural Network in Cybersecurity

and the Internet of Things . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Will Serrano

Autonomous Vehicles - Aerial Vehicles

A Visual Neural Network for Robust Collision Perception in Vehicle

Driving Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Qinbing Fu, Nicola Bellotto, Huatian Wang, F. Claire Rind,
Hongxin Wang, and Shigang Yue

An LGMD Based Competitive Collision Avoidance Strategy for UAV . . . . . 80

Jiannan Zhao, Xingzao Ma, Qinbing Fu, Cheng Hu, and Shigang Yue

Mixture Modules Based Intelligent Control System

for Autonomous Driving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Tangyike Zhang, Songyi Zhang, Yu Chen, Chao Xia, Shitao Chen,
and Nanning Zheng

Biomedical AI

An Adaptive Temporal-Causal Network Model for Stress Extinction

Using Fluoxetine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
S. Sahand Mohammadi Ziabari
xx Contents

Clustering Diagnostic Profiles of Patients . . . . . . . . . . . . . . . . . . . . . . . . . . 120

Jaakko Hollmén and Panagiotis Papapetrou

Emotion Analysis in Hospital Bedside Infotainment Platforms Using

Speeded up Robust Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
A. Kallipolitis, M. Galliakis, A. Menychtas, and I. Maglogiannis

FISUL: A Framework for Detecting Adverse Drug Events from

Heterogeneous Medical Sources Using Feature Importance. . . . . . . . . . . . . . 139
Corinne G. Allaart, Lena Mondrejevski, and Panagiotis Papapetrou

Classification - Clustering

A New Topology-Preserving Distance Metric with Applications

to Multi-dimensional Data Clustering. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Konstantinos K. Delibasis

Classification of Incomplete Data Using Autoencoder

and Evidential Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Suvra Jyoti Choudhury and Nikhil R. Pal

Dynamic Reliable Voting in Ensemble Learning . . . . . . . . . . . . . . . . . . . . . 178

Agus Budi Raharjo and Mohamed Quafafou

Extracting Action Sensitive Features to Facilitate Weakly-Supervised

Action Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Zijian Kang, Le Wang, Ziyi Liu, Qilin Zhang, and Nanning Zheng

Image Recognition Based on Combined Filters with Pseudoinverse

Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Xiaodan Deng, Xiaoxuan Sun, Ping Guo, and Qian Yin

Constraint Programming - Brain Inspired Modeling

Design-Parameters Optimization of a Deep-Groove Ball Bearing

for Different Boundary Dimensions, Employing Amended Differential
Evolution Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Parthiv B. Rana, Jigar L. Patel, and D. I. Lalwani

Exploring Brain Effective Connectivity in Visual Perception

Using a Hierarchical Correlation Network . . . . . . . . . . . . . . . . . . . . . . . . . 223
Siyu Yu, Nanning Zheng, Hao Wu, Ming Du, and Badong Chen

Solving the Talent Scheduling Problem by Parallel

Constraint Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Ke Liu, Sven Löffler, and Petra Hofstedt
 Contents xxi

Deep Learning - Convolutional ANN

A Deep Reinforcement Learning Approach for Automated

Cryptocurrency Trading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Giorgio Lucarelli and Matteo Borrotti

Capacity Requirements Planning for Production Companies

Using Deep Reinforcement Learning: Use Case for Deep Planning
Methodology (DPM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Harald Schallner

Comparison of Neural Network Optimizers for Relative Ranking Retention

Between Neural Architectures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
George Kyriakides and Konstantinos Margaritis

Detecting Violent Robberies in CCTV Videos Using Deep Learning . . . . . . . 282

Giorgio Morales, Itamar Salazar-Reque, Joel Telles, and Daniel Díaz

Diversity Regularized Adversarial Deep Learning . . . . . . . . . . . . . . . . . . . . 292

Babajide O. Ayinde, Keishin Nishihama, and Jacek M. Zurada

Interpretability of a Deep Learning Model for Rodents Brain

Semantic Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Leonardo Nogueira Matos, Mariana Fontainhas Rodrigues,
Ricardo Magalhães, Victor Alves, and Paulo Novais

Learning and Detecting Stuttering Disorders . . . . . . . . . . . . . . . . . . . . . . . . 319

Fabio Fassetti, Ilaria Fassetti, and Simona Nisticò

Localization of Epileptic Foci by Using Convolutional Neural Network

Based on iEEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
Linfeng Sui, Xuyang Zhao, Qibin Zhao, Toshihisa Tanaka,
and Jianting Cao

Review Spam Detection Using Word Embeddings

and Deep Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
Aliaksandr Barushka and Petr Hajek

Tools for Semi-automatic Preparation of Training Data for OCR. . . . . . . . . . 351

Ladislav Lenc, Jiří Martínek, and Pavel Král

Training Strategies for OCR Systems for Historical Documents . . . . . . . . . . 362

Jiří Martínek, Ladislav Lenc, and Pavel Král

A Review on the Application of Deep Learning in Legal Domain . . . . . . . . . 374

Neha Bansal, Arun Sharma, and R. K. Singh
xxii Contents

Long-Short Term Memory for an Effective Short-Term Weather

Forecasting Model Using Surface Weather Data . . . . . . . . . . . . . . . . . . . . . 382
Pradeep Hewage, Ardhendu Behera, Marcello Trovati, and Ella Pereira

Segmentation Methods for Image Classification Using a Convolutional

Neural Network on AR-Sandbox. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Andres Ovidio Restrepo Rodriguez, Daniel Esteban Casas Mateus,
Paulo Alonso Gaona Garcia, Adriana Gomez Acosta,
and Carlos Enrique Montenegro Marin

Fuzzy Modeling

A Hybrid Model Based on Fuzzy Rules to Act on the Diagnosed

of Autism in Adults. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
Augusto J. Guimarães, Vinicius J. Silva Araujo, Vanessa S. Araujo,
Lucas O. Batista, and Paulo V. de Campos Souza

An Unsupervised Fuzzy Rule-Based Method for Structure Preserving

Dimensionality Reduction with Prediction Ability . . . . . . . . . . . . . . . . . . . . 413
Suchismita Das and Nikhil R. Pal

Interpretable Fuzzy Rule-Based Systems for Detecting Financial

Statement Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Petr Hajek

Learning Automata - Logic Based Reasoning

Learning Automata-Based Solutions to the Single Elevator Problem . . . . . . . 439

O. Ghaleb and B. John Oommen

Optimizing Self-organizing Lists-on-Lists Using Enhanced

Object Partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
O. Ekaba Bisong and B. John Oommen

EduBAI: An Educational Platform for Logic-Based Reasoning . . . . . . . . . . . 464

Dimitrios Arampatzis, Maria Doulgeraki, Michail Giannoulis,
Evropi Stefanidi, and Theodore Patkos

Machine Learning - Natural Language

A Machine Learning Tool for Interpreting Differences in Cognition Using

Brain Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Tiago Azevedo, Luca Passamonti, Pietro Lió, and Nicola Toschi
 Contents xxiii

Comparison of the Best Parameter Settings in the Creation and Comparison

of Feature Vectors in Distributional Semantic Models Across
Multiple Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
András Dobó and János Csirik

Distributed Community Prediction for Social Graphs Based

on Louvain Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
Christos Makris, Dionisios Pettas, and Georgios Pispirigos

Iliou Machine Learning Data Preprocessing Method for Suicide Prediction

from Family History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
Theodoros Iliou, Georgia Konstantopoulou, Christina Lymperopoulou,
Konstantinos Anastasopoulos, George Anastassopoulos,
Dimitrios Margounakis, and Dimitrios Lymberopoulos

Ontology Population Framework of MAGNETO for Instantiating

Heterogeneous Forensic Data Modalities . . . . . . . . . . . . . . . . . . . . . . . . . . 520
Ernst-Josef Behmer, Krishna Chandramouli, Victor Garrido,
Dirk Mühlenberg, Dennis Müller, Wilmuth Müller, Dirk Pallmer,
Francisco J. Pérez, Tomas Piatrik, and Camilo Vargas

Random Forest Surrogate Models to Support Design Space Exploration

in Aerospace Use-Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
Siva Krishna Dasari, Abbas Cheddad, and Petter Andersson

Stacking Strong Ensembles of Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . 545

Stamatios-Aggelos N. Alexandropoulos, Christos K. Aridas,
Sotiris B. Kotsiantis, and Michael N. Vrahatis

Multi Agent - IoT

An Agent-Based Framework for Complex Networks . . . . . . . . . . . . . . . . . . 559

Alexander Wendt, Maximilian Götzinger, and Thilo Sauter

Studying Emotions at Work Using Agent-Based Modeling and Simulation. . . 571

Hanen Lejmi-Riahi, Mouna Belhaj, and Lamjed Ben Said

Towards an Adaption and Personalisation Solution Based on Multi Agent

System Applied on Serious Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584
Spyridon Blatsios and Ioannis Refanidis

Nature Inspired Flight and Robot Control - Machine Vision

Constant Angular Velocity Regulation for Visually Guided

Terrain Following . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
Huatian Wang, Qinbing Fu, Hongxin Wang, Jigen Peng,
and Shigang Yue
xxiv Contents

Motion Segmentation Based on Structure-Texture Decomposition

and Improved Three Frame Differencing . . . . . . . . . . . . . . . . . . . . . . . . . . 609
Sandeep Singh Sengar

Using Shallow Neural Network Fitting Technique to Improve Calibration

Accuracy of Modeless Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623
Ying Bai and Dali Wang

Recommendation Systems

Banner Personalization for e-Commerce . . . . . . . . . . . . . . . . . . . . . . . . . . . 635

Ioannis Maniadis, Konstantinos N. Vavliakis,
and Andreas L. Symeonidis

Hybrid Data Set Optimization in Recommender Systems Using Fuzzy

T-Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
Antonios Papaleonidas, Elias Pimenidis, and Lazaros Iliadis

MuSIF: A Product Recommendation System Based on Multi-source

Implicit Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660
Ioannis Schoinas and Christos Tjortjis

On the Invariance of the SELU Activation Function on Algorithm

and Hyperparameter Selection in Neural Network Recommenders . . . . . . . . . 673
Flora Sakketou and Nicholas Ampazis

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687

Invited Paper
 The Power of the “Pursuit” Learning
Paradigm in the Partitioning of Data

Abdolreza Shirvani1 and B. John Oommen1,2(B)

1
School of Computer Science, Carleton University, Ottawa, Canada
oommen@scs.carleton.ca
2
Centre for Artificial Intelligence Research, University of Agder,
Grimstad, Norway

Abstract. Traditional Learning Automata (LA) work with the under-

standing that the actions are chosen purely based on the “state” in which
the machine is. This modus operandus completely ignores any estimation
of the Random Environment’s (RE’s) (specified as E) reward/penalty
probabilities. To take these into consideration, Estimator/Pursuit LA
utilize “cheap” estimates of the Environment’s reward probabilities to
make them converge by an order of magnitude faster. This concept is
quite simply the following: Inexpensive estimates of the reward prob-
abilities can be used to rank the actions. Thereafter, when the action
probability vector has to be updated, it is done not on the basis of the
Environment’s response alone, but also based on the ranking of these
estimates. While this phenomenon has been utilized in the field of LA,
until recently, it has not been incorporated into solutions that solve par-
titioning problems. In this paper, we will submit a complete survey of
how the “Pursuit” learning paradigm can be and has been used in Object
Partitioning. The results demonstrate that incorporating this paradigm
can hasten the partitioning by a order of magnitude.

Keywords: Object Partitioning · Learning Automata ·

Object Migration Automaton · Partitioning-based learning

1 Introduction

The Pursuit Concept in LA: Absolutely Expedient LA are absorbing and

there is always a small probability of them not converging to the best action.
Thathachar and Sastry realized this phenomenon and proposed to use Maxi-
mum Likelihood Estimators (MLEs) to hasten the LA’s convergence. Such an
MLE-based update method would utilize estimates of the reward probabilities in
the update equations. At every iteration, the estimated reward vector was also

The second author gratefully acknowledges the partial support of NSERC, the Natural
Sciences and Engineering Council of Canada.
c IFIP International Federation for Information Processing 2019
Published by Springer Nature Switzerland AG 2019
J. MacIntyre et al. (Eds.): AIAI 2019, IFIP AICT 559, pp. 3–16, 2019.
https://doi.org/10.1007/978-3-030-19823-7_1
4 A. Shirvani and B. John Oommen

used to update the action probabilities, instead of updating it based only on the
RE’s feedback. In this way, the probabilities of choosing the actions with higher
reward estimates were increased, and those with lower estimates were signifi-
cantly reduced, using which they proposed the family of estimator algorithms.
The Pursuit strategy of designing LA is a special derivative of the family
of estimator algorithms. Pursuit algorithms “pursue” the currently-known best
action, and increase the action probability associated with this action. The pur-
suit concept was first introduced by Thathachar et al., and the corresponding
LA was proven to be -optimal. Its discretized version was proposed by Lanc-
tot et al. in [7], who also discretized the original estimator algorithm. Agache
et al. [10] then analyzed all the four linear combinations, i.e., the LRI and LRP
paradigms.
The Object Partitioning Problem (OPP): Consider the problem of par-
titioning a set A = {A1 , · · · , AW } of W physical objects into R groups
Ω = {G1 , · · · , GR }. We assume that the true but unknown state of nature,
Ω ∗ , is a partitioning of the set A into mutually exclusive and exhaustive sub-
sets {G∗1 , G∗2 , · · · G∗R }. The composition of {G∗i } is unknown, and the elements in
the subsets fall together based on some criteria which may be mathematically
formulated, or may even be ambiguous. These objects are now presented to a
learning algorithm, for example, in pairs or tuples. The goal of the algorithm is
to partition A into a learned partition, Ω + . The hope is to have Ω + converge
to Ω ∗ . In most cases, the underlying partitioning of Ω ∗ is not known, nor are
the joint access probabilities by which the pairs/tuples of A are presented to the
learning algorithm known. This problem is known to be NP-hard [9]. Clearly,
if we increase the number of objects, the number of partitions increases, and
in addition to this quantity, the problem’s complexity grows exponentially. To
resolve this, it is possible to explore all partition combinations, use a ranking
index, and to thereafter, report the best plausible partition. The goal of the OPP
is to identify the best or most likely realizable partitioning. This requires the AI
algorithm to perceive the semantic physical world aspects of the objects, and
to then make local decisions based on the best partition in the abstract domain
[4,5].
Real versus Abstract Objects: If there exists a mutual relation between
the real objects in the semantic domain A, and a domain of abstract objects
O = {O1 , · · · , OW }, we define the partitioning of O in such way that the cor-
responding partitions of O map onto the partitions of the real objects in A so
as to mimic the state-of-nature. Thus, while we operate on the abstract objects
in O, the objects in A are not necessarily moved because they constitute real-
life objects which cannot be easily moved. A special case of the OPP is the
Equi-Partitioning Problem (EPP) in which all the partitions are equi-sized.
The Object Migrating Automation (OMA): Due to the poor convergence
of prior OPP/EPP solutions, they were never utilized in real-life applications.
The introduction of an LA-based partitioning algorithm, the OMA, (explained
in Sect. 2) made real-life applications possible. The OMA resolved the EPP both
 The Power of the Pursuit Learning Paradigm in the Partitioning of Data 5

efficiently and accurately. This solution is regarded as a benchmark for the EPP.
Indeed, since 19861 , it has been applied to variety of real-life problems and
domains which include keyboard optimization, image retrieval, distributed com-
puting, graph partitioning, the constraint satisfaction problems, cryptanalysis,
reputation systems, parallel and distributed mapping etc.
The Intent of this Paper: Although the “Pursuit” learning paradigm has
been utilized in the theory and applications of LA as fundamental machines,
until recently, it has not been incorporated into solutions that solve partitioning
problems. The goal of this paper is to submit a comprehensive survey of how this
paradigm can be used in Object Partitioning, and to optimize various versions of
the OMA. We also include simulation results on benchmark environments that
demonstrate the advantages of incorporating it into the respective machines.

2 The Object Migration Automata

The OMA is a fixed structure LA designed to solve the EPP. It is defined as

a quintuple with R actions2 , each of which represents a specific class, and for
every action there exist a fixed number of states, N . Every abstract object from
the set O resides in a state identified by a number, and can move from one state
to another, or migrate from one group to another. If the abstract object Oi is in
state ξi belonging to a group αk , we say that Oi is assigned to class k.
If two objects Oi and Oj happen to be in the same class and the OMA receives
a query Ai , Aj , they are jointly rewarded by E, the Environment. Otherwise,
they will be penalized. We formalize the movements of {Oi } on reward/penalty.
We shall formalize the LA as follows: For every action αk , there is a set of
states {φk1 , · · · , φkN }, where N is the fixed depth of the memory, and where
1 ≤ k ≤ R represents the number of desired classes. We also assume that φk1 is
the most internal state and that φkN is the boundary state for the corresponding
action. The response to the reward and penalty feedback are as follows:

– Reward: Given a pair of physical objects presented as a query Ai , Aj , if

both Oi , and Oj happen to be in the same class αk , the reward scenario is
enforced, and they are both moved one step toward the actions’s most internal
state, φk1 . This is depicted in Figure 3.2(a) in [11]3 .
– Penalty: If, however, they are in different classes, αk and αm , (i.e., Oi ,
is in state ξi where ξi ∈ {φk1 , · · · , φkN } and Oj , is in state ξj where ξj ∈
{φm1 , · · · , φmN }) they are moved away from φk1 and φm1 as follows:
1. If ξi = φkN and ξj = φmN , we move Oi and Oj one state toward φkN
and φmN , respectively, as shown in Figure 3.2(b) in [11].

1
The bibliography in this paper is necessarily limited. The majority of the present
results very briefly summarize the results in the Ph.D. thesis of the First Author.
2
To be consistent with the terminology of LA, we use the terms “action”, “class” and
“group” synonymously.
3
The OMA’s algorithms/figures are in [11], and omitted here in the interest of space.
6 A. Shirvani and B. John Oommen

2. If ξi = φkN or ξj = φmN but not both (i.e., only one of these abstract
objects is in the boundry state), the object which is not in the bound-
ary state, say Oi , is moved towards its boundary state as shown in
Figure 3.2(c) in [11]. Simultaneously, the object that is in the boundary
state, Oj , is moved to the boundary state of Oj . Since this reallocation
will result in an excess of objects in αk , we choose one of the objects in
αk (which is not accessed) and move it to the boundary state of αm . In
this case, we choose the object nearest to the boundary state of ξi , as
shown in Figure 3.2(c) in [11].
3. If ξi = φkN and ξj = φmN (both objects are in the boundary states),
one object, say Oi , will be moved to the boundary state of αm . Since this
reallocation, will again, result in an excess of objects in αm , we choose one
of the objects in αm (which is not accessed) and move it to the boundary
state of αk . In this case, we choose the object nearest to the boundary
state of ξj , as shown in Figure 3.2(d) in [11].
To asses the partitioning accuracy and the convergence speed of any EPP
solution, there must be an “oracle” with a pre-defined number of classes, and with
each class containing an equal number of objects. The OMA’s goal is to migrate
the objects between its classes, using the incoming queries. E is characterized by
three parameters: (a) W , the number of objects, (b) R, the number of partitions,
and (c) a probability ‘p’ quantifying how E pairs the elements in the query.

Table 1. Experimental results for the OMA done for an ensemble of 100 experiments
in which we have only included the results from experiments where convergence has
occurred.

W W/R R OMAp9 OMAp8 OMAp7

4 2 2 (2, 26) (2, 36) (2, 57)
6 2 3 (3, 44) (4, 62) (4, 109)
- 3 3 (22, 66) (20, 88) (26, 153)
9 3 3 (44, 110) (43, 144) (70, 261)
12 2 6 (10, 101) (12, 146) (15, 285)
- 3 4 (82, 172) (84, 228) (128, 406)
- 4 3 (401, 524) (252, 405) (256, 552)
- 6 2 (2240, 2370) (1151, 1299) (1053, 1486)
15 3 5 (152, 265) (155, 325) (191, 607)
- 5 3 (1854, 2087) (918, 1136) (735, 1171)
18 2 9 (17, 167) (24, 252) (29, 582)
- 3 6 (180, 319) (202, 413) (288, 839)
- 6 3 (5660, 5786) (1911, 2265) (1355, 2111)
- 9 2 (11245, 11456) (6494, 7016) (3801, 4450)
 The Power of the Pursuit Learning Paradigm in the Partitioning of Data 7

Every query presented to the OMA by E consists of two objects. E randomly

selects an initial class with probability R1 , and it then chooses the first object in
the query from it, say, q1 . The second element of the pair, q2 , is then chosen with
the probability p from the same class, and with the probability (1 − p) from one
of the other classes uniformly, each of them being chosen with the probability of
1
R−1 . Thereafter, it chooses a random element from the second class uniformly.
We assume that E generates an “unending” continuous stream of query pairs.
The results of the simulations are given in Table 1, where in OM ApX , X
refers to the probability specified above, W , is the number of objects, W/R is
the number of objects per class, and R is the number of classes. The results are
given as a pair (a, b) where a refers to the number of iterations for the OMA to
reach the first correct classification and b refers to the case where the OMA has
fully converged. In all experiments, the number of states of the OMA is set to
10. Also, the OMA’s convergence for a single run and for an ensemble of runs
display a monotonically decreasing pattern (with time) for the latter.

3 Developing the Pursuit Concept: The Environment

In an “un-noisy” Environment, we can denote the actual value of the relation
between Ai and Aj (for k ∈ {1, · · · , R}) by the quantity μ∗ (i, j), expressed as:

μ∗ (i, j) = P (Rk ) · P (Aj |Ai ) · P (Ai ), ∀i, j if Ai , Aj  ∈ RK ,

= 0 otherwise,

where P (Rk ) is the probability that the first element, Ai , is chosen from the
group Rk , and P (Aj |Ai ) is the conditional probability of choosing Aj , which
is also from Rk , after Ai has been chosen. Since E chooses the elements of the
pairs from the other groups uniformly, with a possible re-numbering operation,
the matrix M∗ = [μ∗ (i, j)] is a block-diagonal matrix given by Eq. (1).
⎡M∗ 0 . . . 0 ⎤
1
⎢ .. ⎥
⎢ 0 M∗2 . ⎥
∗ ⎢
M =⎣ .. . ⎥
. .. ⎦
.. (1)
.
0 · · · . . . M∗R

where 0 represents a square matrix containing only 0’s.

Theorem 1. The matrix M∗r , (1 ≤ r ≤ R), is a matrix of probabilities of size
R × R possessing the following form:
W W

⎡ ⎤
0 R
W (W −R) ··· R
W (W −R)
⎢ R
··· R ⎥
⎢ W (W −R) 0 W (W −R) ⎥
∗
Mr = ⎢ ⎢ ⎥ (2)
.. .. .. ⎥
⎣ . . . ⎦
R
W (W −R) ··· R
W (W −R) 0
Another random document with
no related content on Scribd:
water on it and measure the heat imparted to the water in a given
time, also the input to the heating element in the same time, from
which data the efficiency may be calculated. In the case of an
electric iron, dampened cloths may be ironed and the actual water
evaporated by the iron, determined by weighing the cloths before
and after the ironing, together with the increase in weight of the cloth
on the ironing board, the time the iron is in use and the temperature
of the cloths. The actual water evaporated is the difference in the
weight of the cloths before and after ironing, minus the increase in
weight of the cloth on the ironing board, which takes up some of the
moisture from the cloths being ironed.
 Earthen Mustard Pots Used as Acid Jars

A Bottle Made from an Earthen Mustard Pot to Contain Acid

A small earthen mustard pot of the type shown makes an ideal

acid pot for the bench, as it is not only acid-proof but will not upset
so easily as the ordinary acid bottle. The large cork, or stopper of
soft wood, thoroughly boiled in hot paraffin, is bored for the insertion
of another paraffined cork holding the acid-brush handle. If a coat of
paraffin is given the handle, it will easily resist the action of the acid
and last much longer.
 Squeezing Paste from Tubes

Tubes of paste, glue, etc., may be more easily handled by

applying an ordinary key, such as found on most cans containing fish
put up in oil. The end of the tube is inserted in the slot of the key and
then turned.—Contributed by J. H. Priestly, Lawrence, Mass.
 Seeing an Alternating Current in a Mirror
It will almost appear impossible to those unfamiliar with laboratory
methods that one may watch the vibrations—3,600 per minute—of
an alternating current in a little pocket mirror without the use of any
apparatus other than a telephone receiver. The experiment is very
interesting and instructive, one that may be performed at practically
no expense.

The Alternations of the Current may be Seen by Looking in the Mirror

Take an ordinary inexpensive watchcase receiver, drill a hole in

the cover for a short piece of brass tubing, to make a gas
connection, and then plug up the center opening with a cork, into
which is tightly fitted a piece of ¹⁄₈-in. tubing. The upper end of this
should be closed with a plug having a central opening about the size
of a pin. Procure a small rectangular pocket mirror and remove the
celluloid covering, and then, across the back, solder a piece of
straight wire to form a vertical spindle, about which the mirror may be
rotated. Connect any resistance, such as a magnet coil of 10 or 20
ohms, in series with an incandescent lamp, and then connect the
receiver terminals to the ends of this resistance. In this manner an
ideal alternating-current supply of a few volts to operate the receiver
safely is secured. Turn on the gas only sufficient to produce a narrow
pencil of flame, not over 1 in. long. Mount the mirror as shown, or
hold the spindle between the thumb and forefinger of the left hand
while rocking it back and forth with the right. Ordinarily only a streak
of light will appear, but immediately upon turning on the current this
streak will be broken up into a series of regular waves, flatter or
sharper according to the speed with which the mirror is rocked. After
carefully noting the wave form, connect the receiver with the primary
of an ordinary medical coil, across the make-and-break, and note the
marked difference in the waves.
By replacing the receiver with a block of wood having a circular
depression, about 2 in. in diameter and ¹⁄₈ in deep, over which is
pasted a disk of smooth paper, the waves set up by the human voice
may be observed if the talking is done loudly and close to the disk.
The gas connection in this case is made from the back of the block,
as shown. As the several vowels are sounded, the characteristic
wave from each will be seen in the mirror. It is also interesting to
increase the pitch of the voice and note how much finer the waves
become.
 Homemade Screen-Door Check

Air-Cushion Check Made of a Bicycle Foot Pump for a Screen Door

An outside screen door causes considerable annoyance by

slamming when exposed to the wind, even if it is equipped with a
bumper. Nothing short of a door check will prevent this slamming, so
I made a very simple pneumatic check for our door, which works
entirely satisfactorily.
A discarded bicycle foot pump was procured and hinged to the
casing over the door, as shown in the illustration. The hinge was
made as follows: Two holes, A, were drilled through the stirrup, as
near the foot plate as possible; two ordinary screw eyes were turned
into the door casing at B, and two pins were passed through the
holes in the screw eyes and the holes in the stirrup. This allows the
pump to swing when the door is opened. The end of the plunger rod
C is flattened and a hole drilled through it to receive the pin at the top
of the bracket D, which is screwed firmly to the door.
The action of the pump when the door is opened can be readily
understood. The check is adjusted very easily by the machine screw
E, which controls the exhaust of the air when the door closes. The
screw is turned into the hole in the base of the pump where the pipe
was originally connected. One side of the end of the screw is slightly
flattened to allow a better adjustment. The pump can be quickly
removed by pulling out the upper pin in the hinge part.—Contributed
by M. C. Woodward, San Diego, California.
 Bushing Made of Brass Tip on Cartridge Fuse

In order to fasten a short piece of tubing in a socket which had

become worn to a funnel shape, without having to tap the socket and
put in a threaded bushing, it was fixed as follows: One of the brass
tips on a spent cartridge fuse was cut off and one of its ends filed
tapering. After trimming the fiber lining so that it would fit snugly over
the tube, it was driven home. The combination of brass and fiber
adjusted itself nicely to the shape of the worn socket and made a
tight fit.
 Opening Springs for a Tennis-Racket Clamp

When putting a tennis racket in a press, it is difficult to keep the

press open to let the racket slip in. This can be easily remedied by
simply putting a small coil spring around each of the four bolts, as
shown. This will always open the press when the bolts are loosened.
—Contributed by W. X. Brodnax, Jr., Bethlehem, Pa.
 Magic-Paper Fortune Telling
At outdoor carnivals and fairs there is usually a fortune teller who
uses a glass wand to cause one’s fortune to appear on a pad of
paper. Anyone may perform this trick by observing the following
directions.
Instead of a glass wand use a long, narrow bottle of glass. Dip a
new pen into copper sulphate, diluted with six parts of water, and
write out the “fortune” on a piece of paper. The writing, when dry, will
not be visible. Next procure two corks to fit the bottle. An unprepared
cork is placed in the bottle and the other is pocketed, after hollowing
it out and inserting a small sponge soaked in pure ammonia.
The bottle with the cork is passed out for examination. The cork is
casually placed into the pocket after it is returned by a bystander. A
pad of paper is then proffered and an initial is placed on the pad of
paper by the person whose “fortune” is to be told. The paper is rolled
up, with the prepared side on the inside, and inserted into the glass
bottle. The fumes of ammonia will develop the mysterious message.
The trick can be repeated if several prepared sheets of paper are on
hand, and always proves of interest in a party of young persons.
 Common Mistakes in Model Making
By H. J. GRAY

M odels made as a pastime or for exhibition purposes should

represent correctly the full-sized machine, not only as regards
general design but also in the proportioning of parts, the finish, and
the choice of materials. The satisfaction derived from the possession
of a model is greater when it is truly representative. Study and
careful measurement of the original are necessary to attain this
result, and provide valuable experience in the application of correct
mechanical principles.
The most conspicuous, though perhaps not the most frequent,
errors made by amateurs are in the proportioning of the various
parts. This usually arises from insufficient study of the original
machine, and is often sufficiently glaring to attract the attention even
of a casual observer. The foundation or base of a model stationary
engine, for example, is often painted to resemble brickwork. This is
correct, provided the spaces are proportioned so as to represent
bricks and not three-ton slabs of granite.
Mistakes are made in the selection of pulley wheels, both as
regards the character and the size of the pulley that would be
suitable for the particular purpose.
The “cheese-head” or flat-head machine screw appears to have a
peculiar fascination for the model maker, judging from the frequency
with which it is misplaced. It is only necessary to consider what
would happen in a full-sized machine if such screws were used for
making joints in valve rods, cylinder covers, slide bars, for fixing
bearing caps, and the like, to realize how completely such a defect
mars the appearance of a model to a discriminating eye. Bolts, or, in
some cases, studs and nuts, should be used to give an appearance
of correct workmanship.
 Fig. 1 Fig. 3

Fig. 2 Fig. 4

Details of Correct and Incorrect Practice in Model Making: Fig. 1, Valve

Rod Joined by “Cheese-Head” Screws, Wrong, and Joined by Joint and
Pin; Fig. 2, Bearing Cap Fixed with Flat-Head Machine Screws, Wrong,
and with Studs and Nuts; Fig. 3, Cylinder Cover Fixed with Flat-Head
Machine Screws, Wrong, and with Studs and Nuts; Fig. 4, Representation
of a Brick Foundation, Incorrectly on Side, and Correctly on End

Many novices make a serious mistake in the character of the finish

given to the various parts. This usually results through devoting
insufficient attention to the method of manufacture adopted in
engineering practice. Under the impression that a mottled
appearance gives an ornamental effect, they will make a shaft end
with a scraped finish. To the casual observer there would be nothing
amiss, but a mistake of this kind would offend the trained eye of an
engineer, because it is entirely unrepresentative. The object of
scraping sliding surfaces is to obtain a greater degree of flatness by
removing small inequalities. As the subsequent use of a file would
only undo the work of a scraper, the surface is permitted to remain
mottled, as left by the scraping tool. But the end of an engine shaft is
not a sliding surface, and in engineering practice would be finished in
a lathe.
Nickelplating is often resorted to in order to produce a brilliant and
supposedly pleasing finish to the model of a casting. This is
obviously wrong, for the actual casting—which might weigh tons—
would be painted, and not electroplated.
Locomotive wheels or stacks of polished brass add to the
appearance of a model only in the eyes of the uninitiated. Few
persons would care to risk a railroad journey if the engine had brass
wheels. Iron or steel is the correct material to use. Brass is also often
used instead of iron for cylinders, connecting rods, and starting
levers on models, or for steam pipes, which should be made of steel
or copper.
In certain cases there may be unusual difficulties in using the
correct material for a machine part made to a small scale. It is then
permissible to use other material, provided some attempt is made to
disguise the fact by means of an appropriate finish. Copperplating,
for example, may be used to disguise some other material, if the
parts should properly be made of copper. It is often convenient to
make a model boiler of brass. It should not be polished but bronzed,
to represent the iron or steel plates of a full-size boiler.
 Take-Down Emergency Oars

When Knocked down the Oars Occupy Small Space in a Boat

Owners of sail or power boats will find the take-down oars shown
in the sketch easily made and of value in an emergency far out of
proportion to the space occupied in a boat. A pair of ordinary oars
was cut as shown, and pipe fittings were attached to the ends to
form a detachable joint. When knocked down the oars may be stored
in a seat cupboard, or other convenient place.—Contributed by H. E.
Ward, Kent, Wash.
 How to Make Propeller Blades Quickly
Requiring a number of propeller blades for use in making models
of windmills, and other constructions, I found that I could save much
time and obtain a satisfactory set of propeller blades by using
ordinary shoehorns of the same size. The small ends of the horns
were flattened out so that they could be fastened to pieces of wood
for bearings, and then hammered to the proper shape for cutting the
air, or receiving the force of the wind.
 Bench Stop

Serviceable bench stops may be made by grooving pieces of

maple, or other close-grained, hard wood and fitting strips of clock
spring into them, as shown in the sketch. The pieces must fit the
holes in the bench top snugly, and the spring will then prevent them
from slipping out. The end of the spring fastened to the stop should
be annealed so that a hole for the screw may be drilled into it readily.
—Contributed by Stanley Mythaler, Spring Valley, Minn.
 How to Make a Good Putty
To make a good putty the following formula should be used: Mix
equal parts of firmly ground whiting and white lead with enough
linseed oil to make a thick liquid; add enough commercial putty to
this to make the consistency of regular putty. This putty will not crack
or crumble, and it costs very little to make. If desired, the commercial
putty may be left out and enough whiting added to take up the liquid.
The life of this putty is four times greater than a commercial putty.—
Contributed by L. E. Fetter, Portsmouth, N. H.
Cupboard for Kitchen Utensils
 The illustration shows a style of a cupboard in which kitchen
utensils can be kept in an orderly manner without taking up a great
deal of space. The cupboard is tall and narrow, and the interior face
of each side is scored at even intervals with saw cuts, ¹⁄₄ in. deep. In
the grooves are placed shelves, which are merely squares of
galvanized iron. By placing the shelf in the proper grooves the space
is adapted to the size of the utensil. The small floor space occupied
allows the cupboard to be placed in the part of the kitchen that is
most convenient.