Download Computational Sciences And Artificial Intelligence In Industry New Digital Technologies For Solving Future Societal And Economical Challenges And Automation Science And Engineering 76 Tero Tuo online ebook texxtbook full chapter pdf

Computational Sciences and Artificial
Intelligence in Industry: New Digital

Technologies for Solving Future
Societal and Economical Challenges ...
and Automation: Science and
Engineering, 76) Tero Tuovinen
Visit to download the full and correct content document:
https://ebookmeta.com/product/computational-sciences-and-artificial-intelligence-in-in
dustry-new-digital-technologies-for-solving-future-societal-and-economical-challenges
-and-automation-science-and-engineering-76-tero-tuo/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Artificial Intelligence, Machine Learning, and Data

Science Technologies: Future Impact and Well-Being for
Society 5.0 (Demystifying Technologies for
Computational Excellence) 1st Edition
https://ebookmeta.com/product/artificial-intelligence-machine-
learning-and-data-science-technologies-future-impact-and-well-
being-for-society-5-0-demystifying-technologies-for-
computational-excellence-1st-edition/
Automation and Autonomy: Labour, Capital and Machines

in the Artificial Intelligence Industry Steinhoff
https://ebookmeta.com/product/automation-and-autonomy-labour-
capital-and-machines-in-the-artificial-intelligence-industry-
steinhoff/
Semantic Web Technologies Research and Applications

Computational Intelligence in Engineering Problem
Solving 1st Edition Archana Patel
https://ebookmeta.com/product/semantic-web-technologies-research-
and-applications-computational-intelligence-in-engineering-
problem-solving-1st-edition-archana-patel/
Primary Mathematics 3A Hoerst
https://ebookmeta.com/product/primary-mathematics-3a-hoerst/
Artificial Intelligence and Digital Diplomacy
Challenges and Opportunities 1st Edition Fatima Roumate
https://ebookmeta.com/product/artificial-intelligence-and-
digital-diplomacy-challenges-and-opportunities-1st-edition-
fatima-roumate/
Artificial Intelligence: Technologies, Applications,

and Challenges 1st Edition Lavanya Sharma (Editor)
https://ebookmeta.com/product/artificial-intelligence-
technologies-applications-and-challenges-1st-edition-lavanya-
sharma-editor/
Computational Intelligence and Data Sciences: Paradigms

in Biomedical Engineering 1st Edition Taylor & Francis
Group
https://ebookmeta.com/product/computational-intelligence-and-
data-sciences-paradigms-in-biomedical-engineering-1st-edition-
taylor-francis-group/
Software Engineering Artificial Intelligence Networking

and Parallel Distributed Computing Studies in
Computational Intelligence 1012 Roger Lee (Editor)
https://ebookmeta.com/product/software-engineering-artificial-
intelligence-networking-and-parallel-distributed-computing-
studies-in-computational-intelligence-1012-roger-lee-editor/
Artificial Intelligence and Machine Learning in Public

Healthcare: Opportunities and Societal Impact Kc
Santosh
https://ebookmeta.com/product/artificial-intelligence-and-
machine-learning-in-public-healthcare-opportunities-and-societal-
impact-kc-santosh/
Intelligent Systems, Control and Automation:
Science and Engineering
Tero Tuovinen
Jacques Periaux
Pekka Neittaanmäki Editors
Computational
Sciences and Artificial
Intelligence in
Industry
New Digital Technologies for
Solving Future Societal
and Economical Challenges
Intelligent Systems, Control and Automation:
Science and Engineering
Volume 76
Series Editor
Kimon P. Valavanis, Department of Electrical and Computer Engineering,
University of Denver, Denver, CO, USA
Advisory Editors
P. Antsaklis, University of Notre Dame, IN, USA
P. Borne, Ecole Centrale de Lille, France
R. Carelli, Universidad Nacional de San Juan, Argentina
T. Fukuda, Nagoya University, Japan
N.R. Gans, The University of Texas at Dallas, Richardson, TX, USA
F. Harashima, University of Tokyo, Japan
P. Martinet, Ecole Centrale de Nantes, France
S. Monaco, University La Sapienza, Rome, Italy
R.R. Negenborn, Delft University of Technology, The Netherlands
António Pascoal, Institute for Systems and Robotics, Lisbon, Portugal
G. Schmidt, Technical University of Munich, Germany
T.M. Sobh, University of Bridgeport, CT, USA
C. Tzafestas, National Technical University of Athens, Greece
Intelligent Systems, Control and Automation: Science and Engineering book series
publishes books on scientific, engineering, and technological developments in this
interesting field that borders on so many disciplines and has so many practical
applications: human-like biomechanics, industrial robotics, mobile robotics, service
and social robotics, humanoid robotics, mechatronics, intelligent control, industrial
process control, power systems control, industrial and office automation, unmanned
aviation systems, teleoperation systems, energy systems, transportation systems,
driverless cars, human-robot interaction, computer and control engineering, but also
computational intelligence, neural networks, fuzzy systems, genetic algorithms,
neurofuzzy systems and control, nonlinear dynamics and control, and of course
adaptive, complex and self-organizing systems. This wide range of topics,
approaches, perspectives and applications is reflected in a large readership of
researchers and practitioners in various fields, as well as graduate students who
want to learn more on a given subject.
The series has received an enthusiastic acceptance by the scientific and
engineering community, and is continuously receiving an increasing number of
high-quality proposals from both academia and industry. The current Series Editor
is Kimon Valavanis, University of Denver, Colorado, USA. He is assisted by an
Editorial Advisory Board who help to select the most interesting and cutting edge
manuscripts for the series:
Panos Antsaklis, University of Notre Dame, USA

Stjepan Bogdan, University of Zagreb, Croatia
Alexandre Brandao, UFV, Brazil
Giorgio Guglieri, Politecnico di Torino, Italy
Kostas Kyriakopoulos, National Technical University of Athens, Greece
Rogelio Lozano, University of Technology of Compiegne, France
Anibal Ollero, University of Seville, Spain
Hai-Long Pei, South China University of Technology, China
Tarek Sobh, University of Bridgeport, USA
Springer and Professor Valavanis welcome book ideas from authors. Potential
authors who wish to submit a book proposal should contact Thomas Ditzinger
(thomas.ditzinger@springer.com)
Indexed by SCOPUS, zbMATH, SCImago.
More information about this series at http://www.springer.com/series/6259

Tero Tuovinen Jacques Periaux
• •
Pekka Neittaanmäki
Editors
Computational Sciences
and Artificial Intelligence
in Industry
New Digital Technologies for Solving Future
Societal and Economical Challenges
123
Editors
Tero Tuovinen Jacques Periaux
Faculty of Information Technology CIMNE, International Center for Numerical
University of Jyväskylä Methods in Engineering
Jyväskylä, Finland Barcelona, Spain
School of Technology
JAMK University of Applied Science
Jyväskylä, Finland
Pekka Neittaanmäki
Faculty of Information Technology
University of Jyväskylä
ISSN 2213-8986 ISSN 2213-8994 (electronic)

Intelligent Systems, Control and Automation: Science and Engineering
ISBN 978-3-030-70786-6 ISBN 978-3-030-70787-3 (eBook)
https://doi.org/10.1007/978-3-030-70787-3
© Springer Nature Switzerland AG 2022
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
The CSAI 2019 Conference (Computational Science and AI in Industry: New

Digital Technologies for Solving Future Societal and Economical Challenges) took
place at Jyväskylä, Finland, on June 12–14, 2019.
It is the first of a series on European Conferences on Computational Science and
Artificial Intelligence.
This new series of ECCOMAS Thematic Conferences on CS and AI has been
launched by the Industry Interest Group (IIG) of ECCOMAS, with the main
objectives of strengthening the industrial liaison and developing contacts between
ECCOMAS in the context of EC programs.
Its main objectives have been bringing together academic and small/large
industries experts involved with computational science and artificial intelligence
tools for exploring possibilities and identifying challenges linked to their applica-
tions in industrial and societal sectors.
For some years now, AI has been gaining momentum.
The rapid digitalization of industry and society is bringing opportunities and
challenges to large and small industries, with a target to significantly improving the
well-being of citizens, in reducing their environmental impact (greener transport
with zero emissions is essential and big data seen as a cutting emissions helper), in
increasing safe autonomous urban, air, and surface transport, mobility, in accessing
efficient (big) data tools for medical applications, and epidemics with safer public
healthcare systems.
In fact, applications are multiplying and directly affecting our daily lives with
image and voice recognition, self-driving cars, disease detection, among others.
Many artificial intelligence (AI) strategies dealing with the collection of large sets
of data will help in mastering new challenges, in particular, in modeling and
simulation.
Data is a key competitive advantage in the global AI race and AI will drive and
develop a society of well-being.
A data policy strategy taking into account the AI requirements and how we
should proceed with its development is therefore essential if the European Union
and other countries wish to attain the goals of sovereignty and strategic autonomy.
v
vi Preface
These goals are ambitious but they are necessary steps in the creation of an AI
industry. CSAI 2019 has opened the door to these goals by activating dual net-
working of two majors computational and computer sciences disciplines: compu-
tational science and artificial intelligence. The aim of CSAI was to provide an
overview of the state-of-the-art and the technology trends in the innovative
hybridization of computational methods and digitalization of industrial and societal
applications.
In 1997, an International Conference took place in Tours, France, titled
“Computational Science for the 21st Century”, with a collection of papers reflecting
the “State of the Art” in computational science. Two decades after this event, in
2019, a second wave of “a marriage à la mode” with the digitalization of industry
and society has been launched with “Computational Sciences and Artificial
Intelligence” in Jyväskylä, Finland: CSAI 2019.
The advances in technology and the ever-growing role of digital sensors and
computers in science have led to an exponential growth in the amount and com-
plexity of (big) data that scientists and industries collect. Artificial intelligence
(AI) is able to create added value for all stakeholders, not only for the industry and
users but for society as well. One amazing example where all actors benefit from the
use of AI can be observed in healthcare.
Concurrently, the data algorithmics of computational sciences is already
improving its performance and productivity in several industrial sectors. For
example, the proliferation of sensors installed on complex systems at every level of
industry help both to fine-tune service in real time and to regulate all production
sequences at the highest possible level. While these principles are mainly applied in
the industry, there is no reason why they would not work for society as well.
Data is appearing as a key competitive advantage in the global CS-AI race!
This first book in the series, assembled with the content of ECCOMAS Thematic
Conferences on Applied Sciences, is exploring methodology development and the
application of computational sciences and AI expert systems in the industrial sector,
healthcare, and technology.
The book is addressed to young researchers and engineers in the fields of
computational science and artificial intelligence, ranging from innovative compu-
tational methods to digital machine learning tools and their coupling used for
solving challenging industrial and societal problems.
The content of this volume is organized into four sections with 17 contributions
in the disciplines classified as follows:
Part I Overview
This chapter is focused on the development of computational science, artificial

intelligence, and data science as part of science paradigm development. A short
overview of history is presented and a delay between the development of hardware
and software is noticed. Development has historically been unbalanced, as hardware
development has allowed the growth of computing ability without necessitating the
use of more efficient computing methods. Artificial intelligence terminologies and
Preface vii
derivative technologies are presented, with some challenges related to scientific

computing and artificial intelligence at a practical level. Potential future trends in the
co-development of methodology in the fields are also considered.
Part II Methodology
The main content of Part II includes:

– Novel strategies for data-driven evolutionary in optimization;
– Interaction between two ways of scientific analysis: Mathematical modeling
with scientific computing and artificial intelligence with machine learning;
– Supervised learning with the use of an artificial neural network;
– Topological gradient method with a parsimonious neural network;
– Error indicators of PDEs using machine learning;
– Minimal machine learning for classification and regression problems;
– Fast solution time for large-scale cluster-wise linear regression problems;
– Clustering validation with missing data using distance estimation.
Part III Medical Applications
The main content of Part III includes:

– Comparison of Kubelka-Munk model and stochastic skin reflectance model in
skin cancer diagnostics;
– Unsupervised numerical characterization of borders of malignant skin tumors
from spectral imagery;
– Unsupervised numerical characterization of borders of malignant skin tumors
from spectral imagery;
– Validation of Knee KL-classifying deep neural network using clinical data;
– Predicting future overweight and obesity from childhood growth data.
Part IV Industrial and Technological Applications
The main content of Part IV includes:

– Applications of industrial WSNs connected by IoT on a gas turbine in O&M
programs for offshore wind farms;
– Combined model-order reduction techniques and ANN for data assimilation and
damage detection in structures; assessment through numerical benchmark;
– Wave propagation simulations and convolutional neural network to retrieve
thin-film coatings from hyperspectral images.
The above applications are all areas of world excellence, representing important
challenges in terms of industrial and societal interest, attracting the involvement of
public and private actors, and requiring strong public leadership ready to trigger
advanced transformations.
viii Preface
Perspectives in Industry and Society:
AI of the future will contribute to creating the greatest values for all stakeholders:
industry, services, and users. To achieve this, the industry will have to be con-
nected, and this is already a reality today. Progress is central to such change, which
is made possible by the digital revolution. The Internet of Things, the cloud, and big
data are an integral part of the industry’s transformation as they gain momentum,
affecting every aspect of society.
The world of data-driven algorithms with computational sciences is greatly
improving performance and productivity in several industrial sectors. The prolif-
eration of sensors installed on complex systems found at every level of industry
help both to fine-tune service in real time and to regulate all production sequences at
the highest possible level. All stakeholders in the industry and society sectors are
now up to speed with the digital age. The connected industry is hastening the pace
of transformations that are already underway.
But these perspectives come at a price. All connected objects, soon to be linked
to developments in artificial intelligence, are potential points of entry for digital
attacks. Improved performance has the negative effect of increasing exposure to
cyber-threats. Cyber-attack can penetrate and exploit the logic of networks,
spreading to contaminate the most established industrial systems, such as social
networks and governmental systems.
Cyber-security is a topic of utmost importance since the progress achieved in
connected industry and society is so great that no one can imagine any move back.
The continuity of innovation relies largely on the confidence placed in the data
generated and systems that suspend them. Security is now a major issue for the
sustainability of states, smart territories, and businesses. The rapid increase in
surface and air transport, medicine, and finance sectors, among others, is the most
tangible demonstration of the above issue.
For better or worse, AI is predicted to have a huge impact on the future of
humanity (Steve Pinker, Professor of Psychology, Harvard University,
Enlightenment now: The case for Reason, Science, Humanism and Progress, in
Scientific Foresight Unit, EPRS, Should we fear artificial intelligence? 2018)!
Our sincere acknowledgments go to our invited plenary and semi-plenary
speakers, parallel sessions organizers, round tables organizers and panelists, session
chair (wo)men, and the contributors of this volume. Offering such a high scientific
and industrial quality level of the CSAI 2019 event would not have been achieved
without their expertise in the AI fields.
We would also like to finally express our gratitude to Dr. Mayra Castro, Senior
Editor, and Prof. Kimon Valavanis, the editor of Intelligent Systems, Control and
Automation: Science and Engineering, for their kind agreement to publish the latest
advanced results in the above series.
Preface ix
In conclusion to this preface, let us quote a say of the internationally renowned

painter, Vincent van Gogh: “Great things are done by a series of small things
brought together”. We hope that the small things assembled in the CSAI 2019
volume will contribute to great things for the future of industry and society!
Jyväskylä, Finland Tero Tuovinen

December 2020 Jacques Periaux
Pekka Neittaanmäki
Contents
Part I Overview
Co-development of Methodology, Applications, and Hardware
in Computational Science and Artificial Intelligence . . . . . . . . . . . . . . . 3
Pekka Neittaanmäki, Matti Savonen, Jacques Periaux, and Tero Tuovinen
Part II Methodology
Novel Strategies for Data-Driven Evolutionary Optimization . . . . . . . . . 11
Swagata Roy and Nirupam Chakraborti
Artificial Intelligence and Computational Science . . . . . . . . . . . . . . . . . . 27
Pekka Neittaanmäki and Sergey Repin
Supervised Learning and Applied Mathematics . . . . . . . . . . . . . . . . . . . 37
Olivier Pironneau
Application of the Topological Gradient to Parsimonious Neural
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Kateryna Bashtova, Mathieu Causse, Cameron James, Florent Masmoudi,
Mohamed Masmoudi, Houcine Turki, and Joshua Wolff
Generation of Error Indicators for Partial Differential Equations
by Machine Learning Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Alexey Muzalevskiy, Pekka Neittaanmäki, and Sergey Repin
Newton Method for Minimal Learning Machine . . . . . . . . . . . . . . . . . . 97
Joonas Hämäläinen and Tommi Kärkkäinen
Limited Memory Bundle Method for Clusterwise Linear
Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Napsu Karmitsa, Adil M. Bagirov, Sona Taheri, and Kaisa Joki
xi
xii Contents
Improving Clustering and Cluster Validation with Missing Data

Using Distance Estimation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Marko Niemelä and Tommi Kärkkäinen
Part III Medical Applications

Kubelka–Munk Model and Stochastic Model Comparison in Skin
Physical Parameter Retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Leevi Annala and Ilkka Pölönen
Unsupervised Numerical Characterization in Determining the Borders
of Malignant Skin Tumors from Spectral Imagery . . . . . . . . . . . . . . . . 153
Ilkka Pölönen, Tero Tuovinen, Hannu-Heikki Puupponen,
Mari Salmivuori, Mari Grönroos, and Noora Neittaanmäki
Validation of Knee KL-classifying Deep Neural Network with Finnish
Patient Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Esko Niinimäki, Juha Paloneva, Ilkka Pölönen, Ari Heinonen,
and Sami Äyrämö
Predicting Future Overweight and Obesity from Childhood Growth
Data: A Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Ilkka Rautiainen, Jukka-Pekka Kauppi, Toni Ruohonen, Eero Karhu,
Keijo Lukkarinen, and Sami Äyrämö
Predicting Overweight and Obesity in Later Life from Childhood
Data: A Review of Predictive Modeling Approaches . . . . . . . . . . . . . . . 203
Ilkka Rautiainen and Sami Äyrämö
Part IV Industrial and Technological Applications

Applications of Industrial IoT and WSNs in O&M Programmes
for Offshore Wind Farms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Sean Loughney and Ariel J. Edesess
Combined Model Order Reduction Techniques and Artificial Neural
Network for Data Assimilation and Damage Detection in Structures . . . 247
Luca Rosafalco, Andrea Manzoni, Stefano Mariani,
and Alberto Corigliano
Using Wave Propagation Simulations and Convolutional Neural
Networks to Retrieve Thin Film Thickness from Hyperspectral
Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Anna-Leena Erkkilä, Jukka Räbinä, Ilkka Pölönen, Timo Sajavaara,
Esa Alakoski, and Tero Tuovinen
Contributors
Esa Alakoski JAMK University of Applied Science, Jyväskylä, Finland

Leevi Annala Faculty of Information Technology, University of Jyväskylä,
Sami Äyrämö Faculty of Information Technology, University of Jyväskylä,
Adil M. Bagirov School of Science, Engineering and Information Technology,
Federation University Australia, Ballarat, VIC, Australia
Kateryna Bashtova Adagos, Ramonville-Saint-Agne, France
Mathieu Causse Adagos, Ramonville-Saint-Agne, France
Nirupam Chakraborti Department of Metallurgical and Materials Engineering,
Indian Institute of Technology, Kharagpur, India
Alberto Corigliano Dipartimento di Ingegneria Civile e Ambientale, Politecnico
di Milano, Milan, Italy
Ariel J. Edesess Department of the Built Environment, Low Carbon Eco-
Innovatory, Liverpool John Moores University, Liverpool, UK
Anna-Leena Erkkilä Faculty of Information Technology, University of
Jyväskylä, Jyväskylä, Finland
Mari Grönroos Joint Authority for Päijät-Häme Social and Health Care,
Dermatology and allergology, Lahti, Finland
Joonas Hämäläinen Faculty of Information Technology, University of Jyväskylä,
Ari Heinonen Faculty of Sport and Health Sciences, University of Jyväskylä,
xiii
xiv Contributors
Cameron James Adagos, Ramonville-Saint-Agne, France;

University of Sheffield, Sheffield, United Kingdom
Kaisa Joki Department of Mathematics and Statistics, University of Turku, Turku,
Finland
Eero Karhu Town of Äänekoski, Äänekoski, Finland
Tommi Kärkkäinen Faculty of Information Technology, University of Jyväskylä,
Napsu Karmitsa Department of Mathematics and Statistics, University of Turku,
Turku, Finland
Jukka-Pekka Kauppi Faculty of Information Technology, University of
Sean Loughney School of Engineering, Liverpool Logistics Offshore and Marine
(LOOM) Research Institute, Liverpool John Moores University, Liverpool, UK
Keijo Lukkarinen Town of Äänekoski, Äänekoski, Finland
Andrea Manzoni MOX, Dipartimento di Matematica, Politecnico di Milano,
Milan, Italy
Stefano Mariani Dipartimento di Ingegneria Civile e Ambientale, Politecnico di
Milano, Milan, Italy
Florent Masmoudi Adagos, Ramonville-Saint-Agne, France
Mohamed Masmoudi Adagos, Ramonville-Saint-Agne, France
Alexey Muzalevskiy Peter the Great St. Petersburg Polytechnic University, St.
Petersburg, Polytechnicheskaya, Russia
Noora Neittaanmäki Department of Pathology, Institute of Biomedicine and
Department of Dermatology and Venereology, Institute of Clinical Sciences,
Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden
Pekka Neittaanmäki Faculty of Information Technology, University of
Marko Niemelä Faculty of Information Technology, University of Jyväskylä,
Esko Niinimäki Faculty of Information Technology, University of Jyväskylä,
Juha Paloneva Department of Orthopaedics, Central Finland Central Hospital,
Jacques Periaux International Centre for Numerical Methods in Engineering
(CIMNE), Barcelona, Spain
Contributors xv
Olivier Pironneau LJLL, Sorbonne Université, Paris, France

Ilkka Pölönen Faculty of Information Technology, University of Jyväskylä,
Hannu-Heikki Puupponen Airbus DS Finland Oy, Jyväskylä, Finland
Jukka Räbinä Faculty of Information Technology, University of Jyväskylä,
Ilkka Rautiainen Faculty of Information Technology, University of Jyväskylä,
Sergey Repin Faculty of Information Technology, University of Jyväskylä,
Luca Rosafalco Dipartimento di Ingegneria Civile e Ambientale, Politecnico di
Milano, Milan, Italy
Swagata Roy Department of Metallurgical and Materials Engineering, Indian
Institute of Technology, Kharagpur, India
Toni Ruohonen Faculty of Information Technology, University of Jyväskylä,
Timo Sajavaara Department of Physics, University of Jyväskylä, Jyväskylä,
Finland
Mari Salmivuori Joint Authority for Päijät-Häme Social and Health Care,
Dermatology and allergology, Lahti, Finland
Matti Savonen Faculty of Information Technology, University of Jyväskylä,
Sona Taheri Mathematical Sciences, School of Science, RMIT University,
Melbourne, Australia, Melbourne VIC, Australia
Tero Tuovinen Faculty of Information Technology, University of Jyväskylä,
Jyväskylä, Finland;
School of Technology, JAMK University of Applied Science, Jyväskylä, Finland
Houcine Turki Adagos, Ramonville-Saint-Agne, France;
University of Nice, Cedex 2, Nice, France
Joshua Wolff Adagos, Ramonville-Saint-Agne, France;
ENSTA Paris, Palaiseau, France
Part I
Overview
Co-development of Methodology,
Applications, and Hardware in
Computational Science and Artificial
Intelligence
Pekka Neittaanmäki, Matti Savonen, Jacques Periaux, and Tero Tuovinen
Abstract In this introductory chapter of the book, we outline the development of

computational science, artificial intelligence, and data science as part of science
paradigm development. Computational science and artificial intelligence have mainly
been developed separately during the past 60 years. We present a short overview
of the history and notice a delay between the development of hardware and soft-
ware. Development has historically been unbalanced, as hardware development has
allowed the growth of computing ability without necessitating the use of more effi-
cient computing methods. Moreover, we outline artificial intelligence terminologies
and derivative technologies, and represent some challenges related to scientific com-
puting and artificial intelligence at a practical level. We also consider potential future
trends in the co-development of the methodology of the fields.
1 Simultaneous Development of Hardware and Software
Simultaneous development of hardware and software is critically important. Devel-

opment between these fields has been historically unbalanced. We have presented
P. Neittaanmäki · M. Savonen
Faculty of Information Technology, University of Jyväskylä, University of Jyväskylä, P.O. Box
35, 40014 Jyväskylä, Finland
e-mail: pekka.neittaanmaki@jyu.fi
M. Savonen
e-mail: matti.j.savonen@jyu.fi
J. Periaux (B)
International Centre for Numerical Methods in Engineering (CIMNE), Barcelona, Spain
e-mail: jperiaux@gmail.com
T. Tuovinen
Faculty of Information Technology, University of Jyväskylä, Jyväskylä, Finland
e-mail: tero.tuovinen@jyu.fi
School of Technology, JAMK University of Applied Science, Jyväskylä, Finland
© Springer Nature Switzerland AG 2022 3

T. Tuovinen et al. (eds.), Computational Sciences and Artificial Intelligence in Industry,
Intelligent Systems, Control and Automation: Science and Engineering 76,
https://doi.org/10.1007/978-3-030-70787-3_1
4 P. Neittaanmäki et al.
Fig. 1 Evolution of various scientific paradigms: hardware development accelerating scientific

development
the historical progression of selected scientific paradigms (see Fig. 1). In Fig. 1,
paradigms are presented in correlation with hardware and methodology develop-
ments that have enabled them. It is easy to observe the progression from empirical
to theoretical, from theoretical to computational, from computational to machine
learning/artificial intelligence (AI) and, finally, to data science/big data analysis. It
is possible to confidently predict that the introduction of functional quantum com-
puting solutions as part of data science will be one of the next major development
steps.
Neuvo [3] presented the delay and interaction from mathematical theory and mod-
els to practical products through the example of telecommunications (see Fig. 2). The
timeline begins with the Fourier analysis in 1822 and the telegraph in 1844. Figure 2
shows us the interaction between theory and application and culminates in the first
GSM call. Telecommunications have since progressed to a new era and the applica-
tions of mobile technology have changed the world we live in. Similar world-defining
developments can be identified by creating timelines starting with, for example, quan-
tum theory and the work of Max Planck. This timeline would prominently feature
such applications as nuclear power, transistors, and lasers. Similarly, we can trace
developments from Albert Einstein’s special and general relativity to satellites and
the mobile communications world of today. In each case, we see a significant delay
from new groundbreaking theoretical work to world-changing applications.
Development of hardware design and scientific computing algorithms and soft-
ware go hand in hand. Namely,
result = algorithm × hardware.
If the used method (algorithm + software) does not utilize the power of hardware
and new technology in the most efficient way possible, only partial potential from the
investments will be returned. This is especially true with new challenges emerging
from modern-day challenges and in a time when we are entering an era of hybridized
computing where quantum computing will be an integral part of problem-solving.
Co-development of Methodology, Applications, and Hardware … 5
Fig. 2 Research progress in telecommunications: advances in physics-related disciplines (top line)

and mathematics (bottom line) (Source Neuvo [3])
2 Co-development Potential of Computational Science and

Artificial Intelligence
The field of artificial intelligence is huge, and its correct classification is a general
topic of discussion. Villani et al. [4] represent that AI is at the crossroads of several
disciplines: computer science, mathematics (logic, optimization, analysis, proba-
bilities, and linear algebra), and cognitive science. The algorithms that underpin it
are based on equally varied approaches: semantic analysis, symbolic representation,
statistical and exploratory learning, and neural networks.
Despite these difficulties, we have outlined our view related to various compo-
nents, terminology, and derivative technologies, which is commonly known as weak
artificial intelligence (see Fig. 3). First, support systems for decision-making, expert
systems, planning, scheduling, optimization, robotics, and computer vision have been
developed.
Thanks to progress in deep learning, we have had major progress in machine
learning techniques:
Deep learning allows the computer to build complex concepts out of simpler concepts.
(Goodfellow et al. [2])
As a result of deep learning advances, natural language processing solutions, like

speech synthetization and transcription, have quickly developed. A similar develop-
ment has occurred in natural language processing. These advancements have allowed
us to increase the value of our data and use more complete data sets.
Fig. 3 Artificial intelligence terminology and derivative technologies
The knowledge of computational sciences should not ignore the knowledge of arti-
ficial intelligence but has to benefit from its new technologies in a win-win alliance.
While the impact of AI can provide a major competitive advantage for businesses
and even for societal issues, computational sciences can provide new and innovative
solutions for problem-solving with algorithms boosted with the use of big data and
AI. Together, these scientific fields have great potential to offer solutions not only to
industrial issues but also to the vast societal challenges of today.
3 Challenges in the Contemporary Field of Scientific

Computing and Artificial Intelligence
Artificial intelligence of the future will make it possible to create the greatest values
for all stakeholders: industry, services, and users. The Internet of Things, Cloud
Computing, and Big Data are an integral part of the industry’s transformation, as
they gain momentum, affecting every aspect of society. In this section, we will
present five challenges in the contemporary field of scientific computing and artificial
intelligence-powered applications.
Unbalanced Development
The first challenge we identify stems from the unbalanced development of computa-
tional methods and hardware. The development of hardware has followed Moore’s
law up until the last few years. Unfortunately, methods (algorithms and software)
Co-development of Methodology, Applications, and Hardware … 7
have developed significantly slower than hardware. Because of the fast hardware
development and cheap price of processors, companies and the research commu-
nity have not had enough interest in methodology development. Furthermore, most
research on energy-efficient computing has revolved around hardware instead of
algorithms and methods.
Understanding the Limitations of the Methodology
The second challenge is that many engineering computations use black box software
and therefore the user is not aware of the limitations on methodology. Very attractive
visualization can give a wrong illusion about the quality of computing. In the scientific
community, the research in the mathematical, statistical, IT- and engineering fields
is separated. Researchers are not aware of the current development in neighbouring
fields.
Time-Criticality
The third challenge is related to the inherent slowness of machine learning algorithms.
They are not suitable for time critical applications, such as cyber attacks, automation
of transportation, and classification of very noisy signals. Noisy signal classification
is required in, for example, neuroscience, space technology development, industrial
IoT, and sensor signal analysis.
Cybersecurity
The fourth challenge is cybersecurity. The world of data algorithms with Computa-
tional Sciences is improving performance and productivity in several industrial sec-
tors. The proliferation of sensors installed on complex systems found at every level
of industry helps both to fine-tune service in real time and to regulate all production
sequences at the highest possible level. This progression includes increasing risks
and requires proactive action. All connected objects are potential points of entry for
digital attacks. Improved performance has a paradoxical effect on increasing expo-
sure to cyber threats. Cyber attacks can exploit network logic to contaminate the most
established industrial systems, such as social networks and governmental systems.
Security is now a major issue for the sustainability of states, smart territories, and
businesses.
Big Data
The fifth challenge is that of the fast-increasing quantity of data. Methodology on
big data has not kept up with the increase in available data. In order to catch up with
the current day quantities of data, we need to increase the effectiveness of computing
and improve the quality of computing results. Moreover, we need new methods of
extracting key features from data sets. Many scientific challenges involve a large
amount of high-dimensional data. Yet, it is known that there are always a small
number of unidentified parameters that encode the crucial part of the data. The
question is how to identify and extract these parameters so that the computing can
be focused on key aspects of the data set.
4 Conclusions
Artificial intelligence of the future will generate the greatest values for all stake-
holders: industry, services, and users. To achieve this, the industry will have to be
connected, and this is already a reality today. Progress is central to such change,
made possible by the digital revolution. The Internet of Things, the Cloud Comput-
ing, and Big Data are an integral part of the industry’s transformation as they gain
momentum, affecting every aspect of society.
The world of data algorithms with Computational Sciences is improving perfor-
mance and productivity in several industrial sectors. All stakeholders in the industry
and society sectors are now up to speed with the digital age. The connected industry
is hastening the pace of transformations that are already underway.
But this comes at a price. All connected objects, soon to be linked to developments
in Artificial Intelligence, are potential points of entry for digital attacks. Cyber attacks
can exploit network logic to contaminate the most established industrial systems. This
is a topic of utmost importance, since the progress achieved in connected industry
and society is so great that no one can imagine any going back.
The pace and continuity of innovation rely largely on the confidence placed in
the data generated and systems that suspend them. Security is now a major issue for
the sustainability of states, smart territories, and businesses. The rapid increase in
surface and air transport, medicine, and finance sectors, among others, is the most
tangible demonstration of the above issue.
Researchers, engineers, and entrepreneurs who contribute to the design, develop-
ment, and commercialization of AI systems will play a decisive role in the digital
society of tomorrow. To ensure this, it is necessary to make them aware, from the start
of their training, of the ethical issues linked to the development of digital technologies
(Villani et al. [4], Bentley et al. [1]).
In this book, the co-development of methodology, applications, and hardware
is considered through the various research studies written by well-known experts
around the world.
References
1. Bentley PJ, Brundage M, Häggström O, Metzinger T (2018) Should we fear artificial intelli-
gence? In-depth analysis. European Union, Brussels
2. Goodfellow I, Bengio Y, Courville A (2016) Deep learning. MIT Press, Cambridge
3. Neuvo Y (2008) Industry needs universities and vice versa. In: Engwall L, Weaire D (eds) The
university in the market: proceedings from a symposium held in Stockholm, 1–3 November
2007. Portland Press, pp 119–126
4. Villani C, Schoenauer M, Bonnet Y, Berthet C, Cornut A-C, Levin F, Rondepierre B (2018) For a
meaningful artificial intelligence: towards a French and European strategy. French Government
Part II
Methodology
Novel Strategies for Data-Driven
Evolutionary Optimization
Swagata Roy and Nirupam Chakraborti
Abstract Novel learning algorithms like Evolutionary Neural Net (EvoNN), Bi-
objective Genetic Programming (BioGP), and Evolutionary Deep Neural Net
(EvoDN2) developed in our laboratory are being widely used in diverse areas of engi-
neering metamodeling and multi-objective optimization of practical interest. These
are intelligent algorithms, based on a nature-inspired approach, trying to mimic some
basic aspects of evolutionary biology in a non-biological context, and follow the prin-
ciples of multi-objective optimization. In this article, the basic working principles of
these algorithms are explained.
1 Introduction
Complex multi-faceted problems are ubiquitous in science and industry, where sev-
eral objectives need to be optimized simultaneously to determine their optimum
trade-offs for the best possible outcomes and benefits. Such problems are not only
complex but also their objectives influence each other. Often a little tinkering with
one may lead to a huge deviation in another. A synchronized optimization of all
the objectives to obtain an optimal set of solutions is defined as multi-objective
optimization problems (MOPs) [4].
Further, these problems come with a feasibility set beyond which solutions have
no practical importance, and there could be some additional constraints as well. These
constraints need to be handled carefully as well. In case of MOPs we usually obtain
a set of Pareto optimal points [4] instead of an unique optimum that we normally
get during a single objective optimization. When two such optimum solutions are
compared, if one solution is better in terms of one objective, it will be inevitably
worse in terms of another. These solutions are called non-dominated to each other.
However, when one solution is better or equal in terms of all the objectives but strictly
S. Roy · N. Chakraborti (B)

Department of Metallurgical and Materials Engineering, Indian Institute of Technology,
Kharagpur 721 302, India
e-mail: nchakrab@metal.iitkgp.ac.in

https://doi.org/10.1007/978-3-030-70787-3_2
12 S. Roy and N. Chakraborti
better in terms of at least one objective, then it is said that this solution is dominating
the other one.
For a multi-objective problem the dominance criterion [4, 6] can be defined math-
ematically. Consider a minimization problem
f i (X ), i = 1, 2, . . . , I (1)
subject to constraints
g j (X ) ≥ 0, j = 1, 2, . . . , J, (2)
where X = {xk | k = 1, 2, . . . , K } is a vector of variables, and i = { f i | i =

1, 2, . . . , I } is the corresponding vector of objectives, then the condition for domi-
nance between any two objective vectors is taken as
(i ≺ m ) ↔ (∀i )( f il ≤ f im ) ∧ (∃m)( f il < f i m) (3)
When neither l nor m dominates the other, they are considered to be non-
dominating to each other.
Evolutionary Algorithms (EA) are one of the techniques which use these criteria
in many fields to achieve optimal solutions successfully [6]. Genetic Algorithms
(GA) are a class of such evolutionary optimization algorithms which are designed
to mimic the biological process of evolution. GAs generally imitate three major
biological processes [4, 6]:
1. selection operation for identifying the candidates for the next generation,
2. crossover operation for the probabilistic exchange of genetic information between
two randomly picked parents, and
3. the mutation operation inducing a small, probabilistic change in the genetic
makeup, resulting in a local search.
Different multi-objective evolutionary algorithms (MOEAs) are constructed based
on the GA strategy [6]. Predator Prey Genetic Algorithm (PPGA) is one such algo-
rithm [15]. PPGA is population based and computes multiple solutions approaching
the Pareto-optimal front simultaneously. Coupling GAs, Neural Networks [30], and
Genetic Programming [25], we have designed a number of training and optimiza-
tion algorithms. The idea is to create optimum metamodels out of non-linear data
containing random noise and to use those models as objectives in a multi-objective
optimization framework. These optimum models should neither overfit nor underfit
the data [5]. The major algorithms developed in our group are named as Evolution-
ary Neural Network (EvoNN) [23, 24], Bi-objective Genetic Algorithm (BioGP) [9],
and Evolutionary Deep Neural Network (EvoDN2) [26], and all of them construct
models by training available data irrespective of problem physics without using any
gradient information. These trained models are further incorporated in MOEAs to
obtain Pareto optimal solutions.
A model that misses the implicit relevant relations in the data is said to have
a high bias and is not very good at predicting the behavior it is modeling. On the
Novel Strategies for Data-Driven Evolutionary Optimization 13
other hand, a model that includes the random noise is said to have a high variance.
While such a model gives good results on the data it is trained on, it fails to make
good predictions in other situations. Hence, a balance between bias and variance is
desired to achieve models that predict the system behavior accurately, while ignoring
the random noise of the dataset. In machine learning and statistics, this is known as
the Bias-Variance Dilemma [6]. In our algorithms, this is simply implemented as the
accuracy versus complexity problem, where a set of optimum models are obtained
through a trade-off between them. The problem of creating surrogate models hence
becomes a bi-objective problem in our approach.
All the three abovementioned algorithms generate models based on this strategy.
EvoNN selects the best model from the non-dominated set of models using Akaike
information criteria (AIC) or the corrected Akaike information criteria (AICc) [1],
while both BioGP and EvoDN2 recommend the model based on the least error or
most accuracy, as their architectures prohibit direct application of those information
criteria. EvoDN2, based on deep neural networks, is a step forward in data-driven
modeling. Both EvoNN and BioGP are not designed to learn from a dataset that is
excessively large.
In recent years, with the advent of successful deep learning, the use of Deep Neural
Networks [29] in learning a large set of data has become feasible. Nowadays, a large
volume of information can be easily obtained from industrial automation systems
that routinely store the process data. Mining relevant information out of these large
piles of datasets needs some very efficient modeling, which deep learning is able
to provide. In the backdrop of stagnation of EvoNN while training some very large
data, EvoDN2 was constructed and so far, it has been successful in dealing with large
datasets with as many as 10,000 inputs [26]. The code can be easily upgraded for
even a larger volume of input data.
All these novel strategies have been rigorously tested using a large number of test
functions [7, 12, 22, 32], and they performed very well. Once the objective functions
are created, BioGP and EvoNN are coded to use PPGA as well as a constraint- based
Reference Vector guided Evolutionary Algorithm (cRVEA) [3] for carrying out the
optimization task, while EvoDN2 uses cRVEA only. In addition, all these algorithms
are capable of highlighting the impact of changing any individual variable on the
objective function, even when the variables are interrelated in a complex manner. This
procedure, Single Variable Response (SVR), is incorporated into all these algorithms.
Major studies using these strategies have been conducted in the materials science
field involving blast furnace process [16, 17], alloy development [26], and even in
optimizing potential parameters for molecular dynamics study [27].
These novel algorithms are discussed in the subsequent sections in detail.
2 Algorithms
Modeling noisy data is a very difficult task, particularly so if the noise is random in
nature. Systematic noise can be filtered easily, but the same cannot be said for random
noise. However, data with random noise are ubiquitous in real-life applications, the
source being various industries and experiments and simulations dealing with large
uncertainty. The major challenge for modeling or metamodeling in such cases is to
avoid the possibilities of overfitting and underfitting as mentioned before. In between
these two extremes somewhere lies the actual model, which the algorithms presented
here attempt to capture based upon the notion of Pareto optimality [18], implemented
in an evolutionary way. The models here are rewarded for their accuracy and penalized
for their parameterization and a set of models showing the best possible trade-offs
between those conflicting requirements ultimately emerge.
As stated before, three modeling strategies are presented here. The Predator-
Prey genetic algorithm (PPGA) [15] is primarily the backbone of BioGP, EvoNN,
and EvoDN2. A population of preys signifies the various models or solutions to
the problem at hand. Weaker preys, i.e., relatively inferior models or solutions, are
exterminated by the predators, which are some artificially introduced entities in the
system for the task of annihilating underperforming solutions following a set of rules.
The predators and preys are placed in a toroidal lattice, in which they are allowed
to roam following certain rules. A concept of neighborhood, akin to what is used in
cellular automata [31], is also introduced. The major steps of PPGA are
1. Define parameters such as the lattice size, number of preys preferred, number of
predators, number of generations, and the probabilities of crossover and mutation.
2. Generate random individuals (models or solutions, as the case may be) and place
them randomly in the lattice.
3. Generate predators with certain fitness criteria based upon objectives and place
the predators randomly. For example, in the case of metamodels, our conflicting
criteria are the error (E) and the complexity (A) of the models. Each predator will
be tested for a fitness value ( f i ) such that for the ith predator,
f i = xi E + (1 − xi )A, xi = 1i . (4)
After the above initialization, the predator-prey model proceeds in the following
steps:
1. Each prey is allowed to move in a random direction, i.e., it gets one of the eight
cells in an 8-cell Moore neighborhood (north, south, east, west, and plus the four
diagonal neighbors), to move into. If the cells they are attempting to move into
are occupied by another prey or predator, then they can try again. Each prey is
allowed 10 such attempts. If the prey is still unable to find a place to move, it
remains where it is.
2. After the preys have moved, they are then allowed to breed. If the prey has no
neighbors, it is not allowed to breed. Otherwise, the prey is allowed to breed with
another randomly selected neighbor to produce an offspring using real crossover
and mutation operators [8]. The offspring is randomly placed anywhere in the
lattice, which can be seen as migration among different clusters of prey across
the solution space. Ten attempts are made to place the child on the lattice. If all
the attempted cells are occupied, the child is discarded.
3. The prey population is under constant threat from the predators, which are initially
allocated at random across the lattice. The predators hunt in series. Selection
pressure is exerted upon the prey population through the predator-prey interaction,
that is, predators are given the task of killing the least-fit prey in their vicinity.
The predators first check their neighborhood to see if there is any prey. If so, the
predator selects the least-fit prey and kills it. The predator then moves onto the
cell held by that prey. If a predator has no neighboring prey, it moves in exactly the
same way as a prey. At the beginning of each generation, the maximum number
of moves for the predators, or for that matter the maximum number of kills is
calculated as
numprey,actual − numprey,prefered
n moves = , (5)
numpredators
where numprey,actual is the number of preys present at the beginning of each gen-
eration and numprey,prefered is the number of preys specified at the very beginning.
item At repeated intervals, a Pareto dominance-based, ranking procedure [4] is
applied and the weaker members of the population are eliminated. This involves a
direct application of the criteria presented in (1)–(3). The process repeats until the
maximum number of generations are attained and then only the non-dominated
members identified through the same ranking strategy are picked up. Such solu-
tions present the best trade-off between the objectives.
2.1 BioGP
Bi-objective Genetic Algorithm (BioGP) [9] had its base on Genetic Programming
(GP) [25]. In GP, tree encoding replaces binary or real data encoding that is more
common in the parlance of genetic and evolutionary computation. Concepts of pop-
ulation, fitness, selection, crossover, and mutation are a bit different in GP due to
this change in environment. GP involves a function set which contains user-defined
mathematical operations like division, square root, etc., and a terminal set where all
the variables and constants are kept. In a binary tree, the operators of the function
set are placed at the nodes, while the members of the terminal set form the leaves.
In recent times, GP has become a highly efficient tool for data-driven modeling
and unlike neural networks, it does not require any pre-defined configuration of
weights, biases, and transfer functions. Thus, it can evolve any mathematical function
representing the system being modeled and can also use the logical conditionals when
needed. In conventional GP, the trees are selected based upon their minimum root
mean square error (RMSE). However, the tree with minimum error often may lead to
overfitting, hence Bi-objective genetic algorithm (BioGP) [9] deals with maximizing
the accuracy of the tree (in other words, minimizing its RMSE) and simultaneously
minimizing its complexity (measured as an average of the depth of the tree and the
number of function nodes used) so that the overfitting can be avoided. This leads to
the bi-objective optimization problem and for that, it follows the PPGA algorithm.
Furthermore, the conventional GP suffers from a problem called bloat [5] where
the trees become very large and become insensitive to any further crossover or muta-
tion. Also, often some sort of execution errors like division by zero happen in a tree,
which would be inevitably quite cumbersome to fix in the case of a large tree. To
circumvent this problem, BioGP grows a number of small sub-trees and agglomerate
them through a set of weights and biases using the Linear Least Square Approach
(LLSQ) [19]. A carefully designed error reduction ratio [5] keeps a tab on the per-
formance of each sub-tree. This prevents trees from growing unmanageably large,
thereby prevents bloat and also ensures mathematically acceptable convergence,
since it is obtained through the LLSQ algorithm instead of GA. Fixing the prob-
lems of any rogue tree also becomes simpler here. Upon convergence, a set of trees
constitute an approximate the best possible trade-offs and among them, as indicated
before, the one with the minimum error is taken as the best possible metamodel and
can be further used as an objective in PPGA or cRVEA algorithms.
An adaptive version of BioGP is also incorporated in a commercial software
named KIMEME developed by Cyber Dyne Srl [11]. The software serves an interac-
tive environment for modeling and optimization of multi-objective problems using
various paradigms including BioGP. A typical KIMEME interface with BioGP is
shown in Fig. 1. In KIMEME, the encoding is done in JAVA, which the users cannot
access or modify. The open-source versions of all the three algorithms presented here
are, however, available in both MATLAB and Python.
Fig. 1 A KIMEME interface with BioGP incorporated

2.2 EvoNN
In the EvoNN algorithm, a population of Artificial Neural Networks (ANN) [23, 24]
of various architecture acts as preys in PPGA. The architecture of the Evolutionary
Neural Network (EvoNN) model is quite simple. ANN consists of nodes, divided
into the input layer (which takes the input from the variables), the output layer (which
gives the output based on the inputs), and one hidden layer. Each node is connected
to one or more nodes above it, i.e., toward the direction of the output node. Each
connection has certain weights attached to it. The more the number of connections,
the more complicated the neural network becomes. The number of active connections
and the values of the weights vary between the members of the population. The hidden
nodes take the values provided by the nodes connected to them, multiplied by the
weight of the connection, and use a transfer function to map an output, which it
forwards to the next layer.
In EvoNN, the lower level connections are optimized using PPGA. The two objec-
tives to be optimized here are the error in the outputs of the model, and the complexity
of the model, measured by counting the number of active connections in the lower
part of the network. However, for using the corrected Akaike Information criteria
(AICc) [1], the weights in both upper and lower layers and also the biases were con-
sidered. A crossover operation, in this case, is defined as swapping of some of the
connections between two neural networks, whereas the mutation operation involves
a change in the value of the weights. For this, the weight to be mutated is pro-
vided with a perturbation through the corresponding weights of two other randomly
selected individuals.
The upper part of the network, i.e., the output layer, uses a linear transfer function
and is optimized using the LLSQ algorithm. This ensures mathematical convergence
of the algorithm. A pseudo-code of EvoNN is shown in Algorithm 1.
Algorithm 1 Pseudocode of EvoNN
begin train
Scale data in between [0, 1]
Generate 2-D toroidal lattice with dimension defined by user
Create a random population of Neural Networks, Prey, defined by
mentioned architecture and nodes in hidden layer
begin PPGA
for each member in Prey do
Deactivate some connections based on a fixed probability
Place member in random locations
end for
Create a population of predators, linearly distributed in [0, 1]
Place predators in random lattice positions
for all generations do
for all layers in Prey do
Calculate standard deviation in chromosome.
end for
Create empty new preys++
for all members in Prey do
Find an empty lattice position in the Moore neighborhood for Prey member
Move Prey member to new location based on probability
end for
Find Prey members in Moore neighborhood
Choose one Prey member
end for
for all layers in Prey member do
Create two offsprings by performing Crossover and Mutation
repeat
Choose random lattice position
if empty then
Place offsprings
Add offsprings to Prey
end if
until offsprings are placed or given 10 tries
end for
Evaluate Error and Complexity
end for
Find rank of Prey members
if kill interval condition satisfied then
Kill Prey with ranks worse than Maximum rank
Create new random population of ANNs for new Preys
end if
for all members in Predators do
Calculate number of predator moves
for number of moves do
Move Predator
end for
end for
for all members in new Prey set do
repeat
Choose random lattice position
if empty then
place member
end if
until members are placed or given 10 tries
end for
end for
end PPGA
Find and save Prey members at Pareto Front
Find Prey member with least AICc
Display and save Training
end train
Fig. 2 A typical EvoDN2 architecture
2.3 EvoDN2
EvoNN is not capable of training a large dataset with data points of 10,000 or more.
Neural Networks with one layer are not enough. For this purpose, EvoNN is aug-
mented to a Deep Neural Networks (DNN) giving rise to the Evolutionary Deep
Neural Network (EvoDN) algorithm [26]. Instead of one hidden layer as in EvoNN,
more than one hidden layer is used here with a flexible number of nodes for each
layer. Similar to EvoNN, connections are made from input variables to the nodes
with random weights, some of which are provided with a zero value to discard those
connections. The last layer is converged using the same LLSQ method. Technically,
the EvoDN code needed some modification in handling the DNN preys compared
to ANN preys of EvoNN. In EvoNN, a 2-D matrix of connections for each prey is
enough, but in EvoDN, with different dimensions of 2-D matrices for each layer, the
concept of cell structure is introduced in the code.
Further modification is done to EvoDN leading to Evolutionary Deep Neural
Network coupled with subnets (EvoDN2) [26]. With this feature, we can divide the
dataset variables into subsets, which are individually passed through deep neural nets
with their individual layers and nodes. At the final layer, these subnets are converged
using LLSQ. This leads to much better training, as it results in high flexibility. The
number of layers for each subnet or how to divide the variable set into subsets are
all defined by users as per preference. Increasing layers or subnets may lead to high
fitting, however, that may also lead to overfitting, and hence PPGA is used to strike
a balance between complexity and accuracy. A schematic of EvoDN2 is shown in
Fig. 2.
For ANNs, the number of connections with non-zero weights in between input and
the hidden nodes is summed up to get the complexity. This is redefined for EvoDN2.
Here, it is taken as

C= n1 [|wi |] , (6)
where C is the complexity, and wi is the matrix of the weights of the connections
between the ith and (i + 1)th layer and n denotes the total number of layers. This
will ignore inactive connections which have no effect on the final output and also will
give certain importance to the connections with larger weights. The mutation and
crossover operators are also modified in EvoDN2 to speed up the computing process
[26]. Unlike EvoNN, EvoDN2 does not use corrected Akaike Information criteria
(AICc) as it is not applicable for this structure with this new definition of complexity.
Here, the model with the least error is chosen. A pseudo-code of EvoDN2 is shown
in Algorithm 2.
Algorithm 2 Pseudocode of EvoDN2

begin train
Define how to divide input variables into subsets;
Define number of layers and nodes for each subnets.
Scale data in between [0, 1]
Generate 2-D toroidal lattice with dimension defined by user
Create a random population of Deep Neural Networks, Prey
Follow PPGA as in EvoNN
Find and save Prey members at Pareto Front
Find Prey member with least error
Display and save Training
end train
2.4 cRVEA
Reference Vector guided Evolutionary Algorithm (RVEA) [3] is a multi-objective

optimization technique that can handle more than three objectives unlike dominance-
based algorithms like PPGA. Due to this reason, though PPGA is used for training
models for BioGP, EvoNN, and EvoDN2, cRVEA is used for the final optimiza-
tion task. RVEA uses a collection of adaptive reference vectors. The convergence
toward the Pareto front occurs along these reference vectors, and the selection occurs
in accordance with an Angular Penalized Distance (APD) defined for this purpose.
Reference Vectors, when initialized, are uniformly distributed in the objective hyper-
sphere and in later generations, they adapt to the instantaneous optimal front. At the
beginning of each generation, the angular distances between each individual and the
reference vectors are calculated, and the individuals are then assigned to their closest
reference vector. Thus, one individual is selected per each reference vector, according
to its APD. This is followed by crossover, mutation, and the adaptation of reference
vectors to the new optimal front. This process is repeated until a termination criterion
is met. A constraint-handling version of this algorithm is generated as well, known
as constraint-based Reference Vector Evolutionary Algorithm (cRVEA) [3], which

has been used in the present case.
3 Test Functions
These algorithms were tested by training models on datasets created using a plethora
of standard testing suites [7, 12, 22, 32]. Each suite comprises a number of opti-
mization problems, specially designed to test various special cases of optimization.
The functions available in these test suites were used to check the efficacy of training
through BioGP, EvoNN, and EvoDN2. The functions in these test problems are fairly
complicated and are often quite difficult to model because of various reasons. The
number of independent variables can range from anywhere between 10 and 30 in the
default configurations of these problems.
Many problems are discrete in the objectives space, and many are non-uniform,
meaning that the density of solutions in the objectives space varies from region to
region. These algorithms performed very well with most of such functions and in the
case of some functions with a huge number of local optima, overfitting was clearly
avoided. The details of these numerical experiments are being uploaded in a public
domain platform [21] and are not repeated here; instead, the gist of a paradigm case
is included. The results of training by EvoNN, BioGP, and EvoDN2 are compared
to see how they fare. Dataset of 100, 1000, and 10000 entries are produced to get an
idea of how these algorithms can train them.
All three training algorithms can cover the entire spectrum of data points quite
well. The spread of the calculated ZDT3 second objective function [7] along with
the same predicted by EvoDN2 are shown in Fig. 3. It was observed that EvoDN2
can predict the entire spectrum for 1000 data points quite well. BioGP and EvoNN
fare similarly for 100 or 1000 data points but for a larger dataset of 10,000 entries,
EvoDN2 is far better than both of them. This can be seen from the predicted second
objective of the ZDT2 function versus calculated by all the three algorithms presented
in Fig. 4. It can be seen that for this large dataset, EvoDN2 can predict best with a
correlation coefficient of ≈0.92, while the corresponding correlation coefficients for
EvoNN and BioGP are ≈0.67 and ≈0.81, respectively. Though both EvoNN and
BioGP work quite nicely for 100 and 1000 data, handling a large set of data using
them is quite difficult. However, EvoDN2 can predict 10,000 data quite adequately.
Thus, the development of EvoDN2 based on deep neural nets proves to be quite
useful.
Fig. 3 Training of ZDT3 second objective by EvoDN2
Fig. 4 Trained second objective of ZDT2 versus calculated by (from top to bottom) a BioGP, b
EvoNN, and c EvoDN2
4 Applications in Materials Research
The real-life applications of extant BioGP and EvoNN are already reported in numer-
ous publications [8, 9, 13, 14, 20, 23, 24]. EvoDN2 is a newer algorithm, its appli-
cations have also begun. Since the details of those specific applications are beyond
the scope of this article, a few are briefly mentioned below.
The application of EvoNN in the bi-objective study of an ironmaking blast furnace
[23] or analysis of leaching data of low-grade manganese ores [24] and that of BioGP
in simulated moving bed processing [9] are already documented. EvoDN2 has been
used in our laboratory to predict optimum mechanical properties of microalloyed
steel [26], to study complex Blast furnace data [16, 17] which involves as many as
eight objectives including the total gas flow in the furnace, coke rate, plate cooling
heat loss, etc., and also [27] in computing the parameters for Modified Embedded
Atom Method (MEAM) potential [2] for Aluminum used in materials design through
molecular dynamics (MD) [10] focused on reducing errors between calculated phys-
ical properties and actual Density Functional Theory (DFT) results by Voter and
Chan [28].
Many of these problems need dealing with more than three objectives where due to
a lack of selection pressure, many common evolutionary algorithms fail. The cRVEA
[3] algorithm, however, can handle such a situation and therefore, it has been used
extensively to optimize them. A single variable response (SVR) procedure is also
embedded in these algorithms [19]. The SVR strategy provides large perturbations
to a single variable, while keeping the others fixed at a base level, thus registering
the effect of that particular variable on the property under study. The application of
these algorithms in the field of materials research is thus listed.
5 Concluding Remarks
The genesis of metamodeling algorithms BioGP, EvoNN, and EvoDN2 is discussed

here. The bi-objective tasks with such metamodels can be handled using the PPGA
algorithms, while for a large number of objectives the cRVEA algorithm works with
sufficient accuracy and is integrated into these data-driven modeling and optimization
schemes. The algorithms were tested using a host of test functions and applied to
a number of real-life problems. The strategy used here is quite generic, and these
algorithms are quite suitable for using extensively in problems of diverse nature
belonging to various disciplines.
References
1. Akaike H (2011) Akaike’s information criterion. In: Lovric M (ed) International encyclopedia
of statistical science. Springer, Berlin
2. Baskes MI (1992) Modified embedded-atom potentials for cubic materials and impurities. Phys
Rev B 46(5):2727–2742
3. Cheng R, Jin Y, Olhofer M, Sendhoff BA (2016) A reference vector guided evolutionary
algorithm for many-objective optimization. IEEE Trans Evol Comput 20(5):773–791
4. Coello Coello CA, Becerra RL (2009) Evolutionary multiobjective optimization in materials
science and engineering. Mater Manuf Process 24(2):119–129
5. Collet P (2007) Genetic programming. In: Rennard J-P (ed) Handbook of research on nature-
inspired computing for economics and management, vol 1, Chapter V. IGI Global, Pennsylva-
nia, pp 59–73
6. Deb K (2001) Multi-objective optimization using evolutionary algorithms. Wiley, Hoboken
7. Deb K, Thiele L, Laumanns M, Zitzler E (2002) Scalable multi-objective optimization test
problems. In: Proceedings of the 2002 congress on evolutionary computation – CEC’02. IEEE,
pp 825–830
8. Forrester A, Sóbester A, Keane A (2008) Engineering design via surrogate modelling: a prac-
tical guide. Wiley, Hoboken
9. Giri BK, Hakanen J, Miettinen K, Chakraborti N (2013) Genetic programming through bi-
objective genetic algorithms with a study of a simulated moving bed process involving multiple
objectives. Appl Soft Comput 13(5):2613–2623
10. Hansson T, Oostenbrink C, van Gunsteren W (2002) Molecular dynamics simulations. Curr
Opin Struct Biol 12(2):190–196
11. Iacca G, Mininno E (2015) Introducing kimeme, a novel platform for multi-disciplinary multi-
objective optimization. In: Rossi F, Mavelli F, Stano P, Caivano D (eds) Advances in artificial
life, evolutionary computation and systems chemistry. 10th Italian Workshop, WIVACE 2015
(Bari, 2015). Springer, Cham, pp 40–52
12. Jiménez F, Gómez-Skarmeta AF, Sánchez G, Deb K (2002) An evolutionary algorithm for
constrained multi-objective optimization. In: Proceedings of the 2002 congress on evolutionary
computation – CEC’02. IEEE, pp 1133–1138
13. Kant A, Suman PK, Giri BK, Tiwari MK, Chatterjee C, Nayak PC, Kumar S (2013) Comparison
of multi-objective evolutionary neural network, adaptive neuro-fuzzy inference system and
bootstrap-based neural network for flood forecasting. Neural Comput Appl 23(1):231–246
14. Kumar Sahu R, Halder C, Sen PK (2016) Optimization of top gas recycle blast furnace emissions
with implications of downstream energy. Steel Res Int 87: 1190–1202. https://doi.org/10.1002/
srin.201500312
15. Li X (2003) A real-coded predator-prey genetic algorithm for multiobjective optimization. In:
2nd international conference on evolutionary multi-criterion optimization (EMO 2003) (Faro,
2003). Proceedings. Springer, Berlin, pp 207–221
16. Mahanta BK, Chakraborti N (2018) Evolutionary data driven modeling and multi objective
optimization of noisy data set in blast furnace iron making process. Steel Res Int 89:1800121
(11 p)
17. Mahanta BK, Chakraborti N (2020) Tri-objective optimization of noisy dataset in blast furnace
iron-making process using evolutionary algorithms. Mater Manuf Process 35(6):677–686
18. Miettinen K (1998) Nonlinear multiobjective optimization. Springer, New York
19. Mondal DN, Sarangi K, Pettersson F, Sen PK, Saxén H, Chakraborti N (2011) Cu-Zn sep-
aration by supported liquid membrane analyzed through multi-objective genetic algorithms.
Hydrometallurgy 107(3–4):112–123
20. Nguyen TN, Siegmund T, Tsutsui W, Liao H, Chen W (2016) Bi-objective optimal design of
a damage-tolerant multifunctional battery system. Mater Design 105:51–65
21. Ojalehto V, Miettinen K (2019) DESDEO: an open framework for interactive multiobjective
optimization. In: Huber S, Geiger MJ, de Almeida AT (eds) Multiple criteria decision making
and aiding: cases on models and methods with computer implementations. Springer, Cham, pp
67–94
22. Osyczka A, Kundu S (1995) A new method to solve generalized multicriteria optimization
problems using the simple genetic algorithm. Struct Optim 10(2):94–99
23. Pettersson F, Chakraborti N, Saxén H (2007) A genetic algorithms based multi-objective neural
net applied to noisy blast furnace data. Appl Soft Comput 7(1):387–397
24. Pettersson F, Biswas A, Sen PK, Saxén H, Chakraborti N (2009) Analyzing leaching data
for low-grade manganese ore using neural nets and multiobjective genetic algorithms. Mater
Manuf Process 24(3):320–330
25. Poli R, Langdon WB, McPhee NF (2008) A field guide to genetic programming. http://www.
gp-field-guide.org.uk
26. Roy S, Saini B, Chakrabarti D, Chakraborti N (2020) Mechanical properties of micro-alloyed
steels studied using an evolutionary deep neural network. Mater Manuf Process 35(6):611–624
27. Roy S, Dutta A, Chakraborti N (2021) A novel method of determining interatomic potential for
Al and Al-Li alloys and studying strength of Al-Al3 Li interphase using evolutionary algorithms.
Comput Mater Sci 190:110258
28. Voter AF, Chen SP (1986) Accurate interatomic potentials for Ni, Al and Ni3 Al. MRS Proc
82:175–180
29. Wason R (2018) Deep learning: evolution and expansion. Cogn Syst Res 52:701–708
30. Wilson B (2014) The machine learning dictionary. http://www.cse.unsw.edu.au/~billw/mldict.
html
31. Wolfram S (1983) Statistical mechanics of cellular automata. Rev Mod Phys 55(3):601–644
32. Zitzler E, Deb K, Thiele L (2000) Comparison of multiobjective evolutionary algorithms:
empirical results. Evol Comput 8(2):173–195
Artificial Intelligence and Computational
Science
Pekka Neittaanmäki and Sergey Repin
Abstract In this note, we discuss the interaction between two ways of scientific
analysis. The first (classical) way is known as Mathematical Modeling (MM). It is
based on a model created by humans and presented in mathematical terms. Scientific
Computing (SC) is an important tool of MM developed to quantitatively analyze the
model. Artificial Intelligence (AI) forms a new way of scientific analysis. AI systems
arise as a result of a different process. Here, we take a sequence of correct input–output
data, perform Machine Learning (ML), and get a model (hidden in a network). In this
process, computational methods are used to create a network type model. We briefly
discuss special methods used for this purpose (such as evolutionary algorithms), give
a concise overview of results related to applications of AI in computer simulation of
real-life problems, and discuss several open problems.
Keywords Scientific computations · Machine learning · Artificial intelligence ·

Evolutionary algorithms
1 Classical Method of Scientific Knowledge
The scientific approach that currently dominates in science and technology is the
result of a long evolution of human knowledge. It has a long history dating back
to antiquity. At the core of this approach is what in modern terminology is called
model. First, models were very simple. They were intuitively motivated and verified
in simple physical experiments (as, e.g., those done in Pisa by Galileo Galilei). The
means of elementary mathematics were quite enough to formalize such models and
their verification was done by direct comparison with experimental data.
As science developed, the corresponding mathematical models became more and
more complex. Being initially algebraic, they later start using differential and inte-
P. Neittaanmäki (B) · S. Repin

Faculty of Information Technology, University of Jyväskylä, 35, FI-40014 Jyväskylä, Finland
e-mail: pekka.neittaanmaki@jyu.fi
S. Repin
e-mail: sergey.repin@jyu.fi
https://doi.org/10.1007/978-3-030-70787-3_3
28 P. Neittaanmäki and S. Repin
gral calculus. Nowadays, mathematical models are typically formed by systems that
include partial differential equations coupled with integral and algebraic equations
and other relations. Mathematical models often give good quantitative results that
are useful for analysis and forecasting, but getting them is impossible without serious
computations based on powerful computers, correct approximation methods, adap-
tive numerical algorithms, and error control. Therefore, the complexity of mathemat-
ical models has generated a new scientific direction, computational science, which is
also called scientific computing (SC). It is an essential part of the classical scientific
method focused on a computational model and analysis of numerical results. We can
define it as follows:
Scientific computing is a multidisciplinary field that uses advanced computing

capabilities to understand and solve complex mathematical problems.
Currently, SC is a huge scientific field, which includes a wide range of sub-

directions. Among the most important, we mention
• Computer simulation methods for various mathematical models;
• Optimization and optimization control;
• Analysis of inverse problems;
• Signal analysis;
• Data-analysis and decision-making support;
• Logistics and combinatoric problems;
• Verification and validation of computer programs;
• Computational finance;
• Computational biology;
• Computational science in engineering;
• Big data and cloud resource management.
It should be outlined that using mathematical models and scientific computations
leads to several difficult problems related to the reliability of the results. In the
recent decades, new methods have been developed to control errors that occur when
replacing a continual mathematical model with a discrete model and errors arising
in the process of computations (see, e.g., [1, 2] and references cited therein). For
well-verified mathematical models and clearly defined data this method of scientific
analysis is quite suitable.
SC is an essential part of the classical way of scientific analysis schematically
presented in Fig. 1. Human is at the center of the scheme. A researcher (or a team of
researchers) creates a Mathematical Model formulated in terms of a specially con-
structed virtual world of notations and symbols called Mathematics. Humans perform
measurements/observations and develop computational methods for the mathemati-
cal model. Also, they compare results obtained in physical and numerical experiments
and modify the model if it is required. In the classical scheme (Fig. 1), computer sim-
ulation methods play an important but subsidiary role, while the most crucial part
Artificial Intelligence and Computational Science 29
Fig. 1 Mathematical modeling
related to the model is completely in the hands of humans. The figure depicts one
cycle of a repeatable process, which starts with an original mathematical model based
on certain theoretical analysis. Certainly, the model can be specified and modified,
but this is exclusively a matter of the researcher. After the model has passed the
necessary verification, it can be used independently so that the Mathematical Model
and Computations replace the Experiment.
The scientific method has been using for many centuries and will undoubtedly be
successfully used further. If we talk about complex and highly intellectual problems,
then at present and in the near future no competitors are foreseen for this modus
operandi. Nevertheless, there are many other important and interesting tasks for the
study (and use) of which a different approach can be applied.
2 Genetic Algorithms
Classical methods of scientific analysis (including mathematics, physics, and other

natural sciences) emerged in the process of technological development, which can
be viewed as a unique way of adaptation invented by humans. All other Earth species
use different adaptation methods focused on a strongly limited amount of problems
essential for survival and expansion. Nervous systems of lower animals are networks
trained to effectively address these basic issues: nutrition and reproduction. Despite
a relatively small size, they cope with the tasks quite effectively.
It is not surprising that the idea of transferring the technology developed by life-
forms to manufactured objects came to many scientists and engineers. A science
called bionics (or biomimetics) studies possible applications of “technological solu-
tions” encompassed in biological objects to various engineering problems (see, e.g.,
[3]). A form of this idea in the application to computational sciences is as follows:
Combine methods of computational mathematics with principles of natural selection.
The first works in this direction have appeared in the mid-60s. They were mostly
concerned with relatively simple optimization problems solved by the so-called evo-
lutionary algorithms (EA) (also called genetic algorithms). These methods use selec-
tion principles analogous to those that regulate the successful evolution of spices (see,
e.g., [4–7]). A genetic algorithm operates with “populations” and generates a new
population by modification of the current population and selection of those “indi-
viduals” that are most acceptable with respect to a certain selection principle. Each
successive population is considered a “new generation”. The selection principle is
defined such that only the “individuals” that satisfy it can be considered as solutions
to the problem studied. Hence, the core of this method is an artificial system of
competitors, which consists of mathematical objects.
There are many different modifications of genetic algorithms that differ in how a
“generation” is changed (from simple stochastic disturbance to crossbreeding of most
successful individuals) and how a new set of individuals is formed. One of the first
areas of application was the problem of minimizing functions, where the criterion for
selection is the value of this function. Despite the large number of different studies of
a complete theory of evolutionary algorithms has not yet been created. In the majority
of cases, the algorithms are purely heuristic, and the results related to convergence
and its rate are very rare. Moreover, EAs are rather slow and cannot compete with
well-known deterministic algorithms (if for the corresponding optimization problem
such an algorithm is applicable).
Sometimes, EAs are considered as a special class of probabilistic optimization
methods, which is worth trying only if no other method can be used. This viewpoint
seems to be too limited. There is no doubt that EAs have a much wider scope than
optimization. From a mathematical point of view, they are closer to discrete dynam-
ical systems and have some typical features, e.g., a sequence of “populations” may
not converge to the desired solution (as in a deterministic algorithm) but fluctuates
around it. The latter behavior is typical for a dynamic system at the vicinity of an
attractor.
The invention of EAs has made an important step toward AI technologies because
they solve complicated problems of very different origin without using deep math-
ematical models. Not being very effective for well-studied classes of optimization
problems, they form a basis for new methods commonly used in machine learning
of networks.
3 AI and New Methods of Scientific Analysis
A strict and commonly accepted definition of artificial intelligence (AI) does not
yet exist. Sometimes, it is understood that AI arises in systems that are able to find
acceptable solutions to complex problems in conditions of incomplete information.
Such types of problems arise in the recognition of images, classification of objects,
decision-making, and many other areas. The ability to analyze and solve such prob-
lems determines the cognitive properties of the system. Therefore, sometimes the
question of the presence or absence of AP in a system is associated with the level of
its cognitive abilities. Biological systems often have a high level of recognition in
areas where it is essential for survival.
Artificial neural networks (ANN) are inspired by ideas of biomimetics. (These
issues are examined in depth and in detail in [8, 9]). However, the question of where
the border between intelligent and non-intelligent systems lies remains open. If we
talk about technical systems, then we can offer the following feature, which allows
us to select a system with AI: it is impossible (or very difficult) to establish why
such a system gave a concrete answer. Moreover, often when reprocessing exactly
the same initial data, the system can give a somewhat different answer. Computing
systems without intellectual properties have completely different properties: they act
according to completely defined rules and always produce the same result for the
same input data.
In modern literature, methods for creating artificial systems with some intelli-
gence (in the above sense) are called machine learning (ML). Image recognition and
inverse problems were the first classes of real-life problems where these methods
have demonstrated high efficiency (see, e.g., [10, 11] and the references cited in these
publications). Recently, similar approaches have been used as new tools of scientific
computing and mathematical modeling, in particular, for getting approximations of
differential equations (see [12–19]), quantitative analysis of energy type mathemat-
ical models in mechanics [20–22], automatic differentiation [23], and optimization
of a robotic system [24].
Here, the principal scheme of data processing and generation of a “model” differs
essentially from the scheme in Fig. 1. In essence, human participation is limited to
three things. They are
1. network structure and its size, i.e., the researcher defines a certain class of models
among which a suitable model must be found (e.g., in many cases, the class of
deep neural networks (DNN) having multiple layers of neurones is used);
2. the quality criterion used for comparing the actual (correct) output data with the
data generated by a network. Typically, the quality criterion is a version of the
least squares principle (see, e.g., [9, 10, 13]);
3. the adaptation or optimization algorithm used in the iterative procedure of net-
work changing (methods of nonlinear programming and structural optimization
are often used as the basic tools).
Fig. 2 Network modeling
In the case of the supervised machine learning method, we also need a sufficiently
wide set of actual data that can be used in the teaching process. Therefore, an impor-
tant component of this technology is the formation of such a set.
The scheme in Fig. 2 presents one cycle of the learning process: generator of input
data. Usually, the learning process occurs fully automatically without human inter-
vention. As a result, we obtain a network, whose structure and weights are adapted
to solve a particular problem. Thus, machine learning creates a model presented in
terms of the network structure.
A comparison of Figs. 1 and 2 shows the fundamental differences between the two
methods of analysis. The main idea of Mathematical Modeling (and its part Scientific
Computing) is Correct mathematical model supplied with a proper computational
method provides the required information about a process or phenomenon. The main
concept of ML is rather different: Using a sufficiently representative set of correct
input–output data, create a network type model of a process or phenomenon.
4 Open Problems and Development Prospect
AI methods are at the very beginning of their development and a unified theory of
them is yet to be created. We believe that such a theory will appear on the crossroad of
discrete mathematics, group theory, representation theory, and computational math-
ematics. Below, we briefly discuss three open problems, which have a fundamental
meaning for the understanding of what is AI and how to create AI systems. To the
best of our knowledge, they are not solved. New results related to the problems could
form a basis of the forthcoming theory of AI.
Problem of Teaching
The learning process is usually related to the optimization of a nonlinear functional
(set of functionals) having large dimensions. Usually, this functional is nonconvex
and has a complex structure with numerous local extrema and stationary points.
Therefore, known minimization algorithms (based on additional assumptions such
as, e.g., convexity or unimodality) may not be efficient enough, and we do not know
how to guarantee that the teaching process used is indeed efficient and generates a
network with the structure close to the best possible (among other networks of the
same size).
Selection of a Suitable Class of Networks
One of the major tasks to be solved first of all is to set the structure and size of the
network, which we are going to train. The problem is how to adjust the topological
structure/parameters of a network to the complexity of the problem in question. It
would be nice to have some quantitative a priori criteria able to define the amount
of layers and neurones depending on the amount of output parameters, variability of
the data, and desired accuracy.
Why a Neural Network Model Works?
If a network has a simple structure and consists of a small amount of elements, then
it is usually possible to understand why it is functioning correctly. However, simple
networks can match only simple problems. Future development of AI technologies
will inevitably lead to very complicated networks that will be created for the analysis
of serious scientific (technological, medical, and social) problems. It is quite probable
that some advanced AI systems will work more effectively than standard methods of
mathematical modeling. In this case, scientists will be interested to understand why
a particular AI system works better than their models. To answer this question, it
is necessary to “decode” the information encompassed in the “black box” structure
and translate it into notations and terms available for humans. This problem does not
attract much attention now but may become one of the fundamental problems in the
close future. We must admit that there is no guarantee that it will always be solvable.
5 Conclusions
Machine learning uses various methods of scientific computing, but as technical

appliances only. In principle, “learning” and generation of an intellectual system can
be performed without SC (as, e.g., in the world of nature). Therefore, AI is principally
not reducible to scientific computing and cannot be viewed as its part. At the same
time, the development of AP methods cannot in any way lead to the fading of interest
in classical methods of computational mathematics.
Another random document with
no related content on Scribd:
the foundation is a very good one, but as I have generally allowed
people to call me what they have pleased, and as there is nothing
necessarily dishonorable in this, I have never taken the pains to
dispute its application and propriety; and yet I confess that I am never
so spoken of without feeling a trifle uncomfortable—about as much so
as when I am called, as I sometimes am, the Rev. Frederick Douglass.
My stay in this legislative body was of short duration. My vocation
abroad left me little time to study the many matters of local legislation;
hence my resignation, and the appointment of my son Lewis to fill out
my term.
I have thus far told my story without copious quotations from my
letters, speeches, or other writings, and shall not depart from this rule
in what remains to be told, except to insert here my speech, delivered
at Arlington, near the monument to the “Unknown Loyal Dead,” on
Decoration Day, 1871. It was delivered under impressive
circumstances, in presence of President Grant, his Cabinet, and a
great multitude of distinguished people, and expresses, as I think, the
true view which should be taken of the great conflict between slavery
and freedom to which it refers.
“Friends and Fellow Citizens: Tarry here for a moment. My

words shall be few and simple. The solemn rites of this hour and
place call for no lengthened speech. There is in the very air of this
resting ground of the unknown dead a silent, subtle, and an all-
pervading eloquence, far more touching, impressive, and thrilling
than living lips have ever uttered. Into the measureless depths of
every loyal soul it is now whispering lessons of all that is precious,
priceless, holiest, and most enduring in human existence.
“Dark and sad will be the hour to this nation when it forgets to
pay grateful homage to its greatest benefactors. The offering we
bring to-day is due alike to the patriot soldiers dead and their noble
comrades who still live; for whether living or dead, whether in time
or eternity, the loyal soldiers who imperiled all for country and
freedom are one and inseparable.
“Those unknown heroes whose whitened bones have been
piously gathered here, and whose green graves we now strew with
sweet and beautiful flowers, choice emblems alike of pure hearts
and brave spirits, reached in their glorious career that last highest
point of nobleness beyond which human power cannot go. They
died for their country.
“No loftier tribute can be paid to the most illustrious of all the
benefactors of mankind than we pay to these unrecognized
soldiers, when we write above their graves this shining epitaph.
“When the dark and vengeful spirit of slavery, always
ambitious, preferring to rule in hell than to serve in heaven, fired
the Southern heart and stirred all the malign elements of discord;
when our great Republic, the hope of freedom and self-
government throughout the world, had reached the point of
supreme peril; when the Union of these States was torn and rent
asunder at the center, and the armies of a gigantic rebellion came
forth with broad blades and bloody hands to destroy the very
foundation of American society, the unknown braves who flung
themselves into the yawning chasm, where cannon roared and
bullets whistled, fought and fell. They died for their country.
“We are sometimes asked, in the name of patriotism, to forget
the merits of this fearful struggle, and to remember with equal
admiration those who struck at the nation’s life and those who
struck to save it,—those who fought for slavery and those who
fought for liberty and justice.
“I am no minister of malice. I would not strike the fallen. I
would not repel the repentant, but may my “right hand forget her
cunning, and my tongue cleave to the roof of my mouth,” if I forget
the difference between the parties to that terrible, protracted, and
bloody conflict.
“If we ought to forget a war which has filled our land with
widows and orphans, which has made stumps of men of the very
flower of our youth; sent them on the journey of life armless,
legless, maimed and mutilated; which has piled up a debt heavier
than a mountain of gold—swept uncounted thousands of men into
bloody graves, and planted agony at a million hearthstones; I say
if this war is to be forgotten, I ask in the name of all things sacred
what shall men remember?
“The essence and significance of our devotions here to-day
are not to be found in the fact that the men whose remains fill
these graves were brave in battle. If we met simply to show our
sense of bravery, we should find enough to kindle admiration on
both sides. In the raging storm of fire and blood, in the fierce
torrent of shot and shell, of sword and bayonet, whether on foot or
on horse, unflinching courage marked the rebel not less than the
loyal soldier.
“But we are not here to applaud manly courage, save as it has
been displayed in a noble cause. We must never forget that victory
to the rebellion meant death to the republic. We must never forget
that the loyal soldiers who rest beneath this sod flung themselves
between the nation and the nation’s destroyers. If to-day we have
a country not boiling in an agony of blood like France; if now we
have a united country, no longer cursed by the hell-black system
of human bondage; if the American name is no longer a by-word
and a hissing to a mocking earth; if the star spangled banner floats
only over free American citizens in every quarter of the land, and
our country has before it a long and glorious career of justice,
liberty, and civilization, we are indebted to the unselfish devotion of
the noble army who rest in these honored graves all around us.”
In the month of April, 1872, I had the honor to attend and preside
over a National Convention of colored citizens, held in New Orleans. It
was a critical period in the history of the Republican party, as well as in
that of the country. Eminent men who had hitherto been looked upon
as the pillars of Republicanism had become dissatisfied with President
Grant’s administration, and determined to defeat his nomination for a
second term. The leaders in this unfortunate revolt were Messrs.
Trumbull, Schurz, Greeley, and Sumner. Mr. Schurz had already
succeeded in destroying the Republican party in the State of Missouri,
and it seemed to be his ambition to be the founder of a new party, and
to him more than to any other man belongs the credit of what was
once known as the Liberal Republican party which made Horace
Greeley its standard bearer in the campaign of that year.
At the time of the Convention in New Orleans the elements of this
new combination were just coming together. The division in the
Republican ranks seemed to be growing deeper and broader every
day. The colored people of the country were much affected by the
threatened disruption, and their leaders were much divided as to the
side upon which they should give their voice and their votes. The
names of Greeley and Sumner, on account of their long and earnest
advocacy of justice and liberty to the blacks, had powerful attractions
for the newly enfranchised class; and there was in this Convention at
New Orleans naturally enough a strong disposition to fraternize with
the new party and follow the lead of their old friends. Against this policy
I exerted whatever influence I possessed, and, I think, succeeded in
holding back that Convention from what I felt sure then would have
been a fatal political blunder, and time has proved the correctness of
that position. My speech on taking the chair on that occasion was
telegraphed from New Orleans in full to the New York Herald, and the
key-note of it was that there was no path out of the Republican party
that did not lead directly into the Democratic party—away from our
friends and directly to our enemies. Happily this Convention pretty
largely agreed with me, and its members have not since regretted that
agreement.
From this Convention onward, until the nomination and election of
Grant and Wilson, I was actively engaged on the stump, a part of the
time in Virginia with Hon. Henry Wilson, in North Carolina with John M.
Longston and John H. Smyth, and in the State of Maine with Senator
Hamlin, Gen. B. F. Butler, Gen. Woodford, and Hon. James G. Blaine.
Since 1872 I have been regularly what my old friend Parker
Pillsbury would call a “field hand” in every important political campaign,
and at each National Convention have sided with what has been called
the stalwart element of the Republican party. It was in the Grant
Presidential campaign that New York took an advanced step in the
renunciation of a timid policy. The Republicans of that State not having
the fear of popular prejudice before their eyes placed my name as an
Elector at large at the head of their Presidential ticket. Considering the
deep-rooted sentiment of the masses against negroes, the noise and
tumult likely to be raised, especially among our adopted citizens of
Irish descent, this was a bold and manly proceeding, and one for which
the Republicans of the State of New York deserve the gratitude of
every colored citizen of the Republic, for it was a blow at popular
prejudice in a quarter where it was capable of making the strongest
resistance. The result proved not only the justice and generosity of the
measure, but its wisdom. The Republicans carried the State by a
majority of fifty thousand over the heads of the Liberal Republican and
the Democratic parties combined.
Equally significant of the turn now taken in the political sentiment of
the country, was the action of the Republican Electoral College at its
meeting in Albany, when it committed to my custody the sealed up
electoral vote of the great State of New York, and commissioned me to
bring that vote to the National Capital. Only a few years before, any
colored man was forbidden by law to carry a United States mail bag
from one post-office to another. He was not allowed to touch the
sacred leather, though locked in “triple steel,” but now, not a mail bag,
but a document which was to decide the Presidential question with all
its momentous interests, was committed to the hands of one of this
despised class; and around him, in the execution of his high trust, was
thrown all the safeguards provided by the Constitution and the laws of
the land. Though I worked hard and long to secure the nomination and
the election of Gen. Grant in 1872, I neither received nor sought office
under him. He was my choice upon grounds altogether free from
selfish or personal considerations. I supported him because he had
done all, and would do all, he could to save not only the country from
ruin, but the emancipated class from oppression and ultimate
destruction; and because Mr. Greeley, with the Democratic party
behind him, would not have the power, even if he had the disposition,
to afford us the needed protection which our peculiar condition
required. I could easily have secured the appointment as Minister to
Hayti, but preferred to urge the claims of my friend, Ebenezer Bassett,
a gentleman and a scholar, and a man well fitted by his good sense
and amiable qualities to fill the position with credit to himself and his
country. It is with a certain degree of pride that I am able to say that my
opinion of the wisdom of sending Mr. Bassett to Hayti has been fully
justified by the creditable manner in which, for eight years, he
discharged the difficult duties of that position; for I have the assurance
of Hon. Hamilton Fish, Secretary of State of the United States, that Mr.
Bassett was a good Minister. In so many words, the ex-Secretary told
me, that he “wished that one-half of his ministers abroad performed
their duties as well as Mr. Bassett.” To those who knew Hon. Hamilton
Fish, this compliment will not be deemed slight, for few men are less
given to exaggeration and are more scrupulously exact in the
observance of law, and in the use of language, than is that gentleman.
While speaking in this strain of complacency in reference to Mr.
Bassett, I take pleasure also in bearing my testimony based upon
knowledge obtained at the State Department, that Mr. John Mercer
Langston, the present Minister to Hayti, has acquitted himself with
equal wisdom and ability to that of Mr. Bassett in the same position.
Having known both these gentlemen in their youth, when the one was
at Yale, and the other at Oberlin College, and witnessed their efforts to
qualify themselves for positions of usefulness, it has afforded me no
limited satisfaction to see them rise in the world. Such men increase
the faith of all in the possibilities of their race, and make it easier for
those who are to come after them.
The unveiling of Lincoln Monument in Lincoln Park, Washington,
April 14th, 1876, and the part taken by me in the ceremonies of that
grand occasion, takes rank among the most interesting incidents of my
life, since it brought me into mental communication with a greater
number of the influential and distinguished men of the country than any
I had before known. There were present the President of the United
States and his Cabinet, Judges of the Supreme Court, the Senate and
House of Representatives, and many thousands of citizens to listen to
my address upon the illustrious man in whose memory the colored
people of the United States had, as a mark of their gratitude, erected
that impressive monument. Occasions like this have done wonders in
the removal of popular prejudice, and in lifting into consideration the
colored race; and I reckon it one of the high privileges of my life, that I
was permitted to have a share in this and several other like
celebrations.
The progress of a nation is sometimes indicated by small things.
When Henry Wilson, an honored Senator and Vice-President of the
United States, died in the capitol of the nation, it was a significant and
telling indication of national advance, when three colored citizens, Mr.
Robert Purvis, Mr. James Wormley, and myself, were selected with the
Senate committee, to accompany his honored remains from
Washington to the grand old commonwealth he loved so well, and
whom in turn she had so greatly loved and honored. It was meet and
right that we should be represented in the long procession that met
those remains in every State between here and Massachusetts, for
Henry Wilson was among the foremost friends of the colored race in
this country, and this was the first time in its history when a colored
man was made a pall-bearer at the funeral, as I was in this instance, of
a Vice-President of the United States.
An appointment to any important and lucrative office under the
United States government, usually brings its recipient a large measure
of praise and congratulation on the one hand, and much abuse and
disparagement on the other; and he may think himself singularly
fortunate if the censure does not exceed the praise. I need not dwell
upon the causes of this extravagance, but I may say there is no office
of any value in the country which is not desired and sought by many
persons equally meritorious and equally deserving. But as only one
person can be appointed to any one office, only one can be pleased,
while many are offended; unhappily, resentment follows
disappointment, and this resentment often finds expression in
disparagement and abuse of the successful man. As in most else I
have said, I borrow this reflection from my own experience.
My appointment as United States Marshal of the District of
Columbia, was in keeping with the rest of my life, as a freeman. It was
an innovation upon long established usage, and opposed to the
general current of sentiment in the community. It came upon the
people of the District as a gross surprise, and almost a punishment;
and provoked something like a scream—I will not say a yell—of
popular displeasure. As soon as I was named by President Hayes for
the place, efforts were made by members of the bar to defeat my
confirmation before the Senate. All sorts of reasons against my
appointment, but the true one, were given, and that was withheld more
from a sense of shame, than from a sense of justice. The
apprehension doubtless was, that if appointed marshal, I would
surround myself with colored deputies, colored bailiffs, colored
messengers, and pack the jury box with colored jurors; in a word,
Africanize the courts. But the most dreadful thing threatened, was a
colored man at the Executive Mansion in white kid gloves, sparrow-
tailed coat, patent leather boots, and alabaster cravat, performing the
ceremony—a very empty one—of introducing the aristocratic citizens
of the republic to the President of the United States. This was
something entirely too much to be borne; and men asked themselves
in view of it, to what is the world coming? and where will these things
stop? Dreadful! Dreadful!
It is creditable to the manliness of the American Senate, that it was
moved by none of these things, and that it lost no time in the matter of
my confirmation. I learn, and believe my information correct, that
foremost among those who supported my confirmation against the
objections made to it, was Hon. Roscoe Conkling of New York. His
speech in executive session is said by the senators who heard it, to
have been one of the most masterly and eloquent ever delivered on
the floor of the Senate; and this too I readily believe, for Mr. Conkling
possesses the ardor and fire of Henry Clay, the subtlety of Calhoun,
and the massive grandeur of Daniel Webster.
The effort to prevent my confirmation having failed, nothing could
be done but to wait for some overt act to justify my removal; and for
this my unfriends had not long to wait. In the course of one or two
months I was invited by a number of citizens of Baltimore to deliver a
lecture in that city in Douglass Hall—a building named in honor of
myself, and devoted to educational purposes. With this invitation I
complied, giving the same lecture which I had two years before
delivered in the city of Washington, and which was at the time
published in full in the newspapers, and very highly commended by
them. The subject of the lecture was, “Our National Capital,” and in it I
said many complimentary things of the city, which were as true as they
were complimentary. I spoke of what it had been in the past, what it
was at that time, and what I thought it destined to become in the future;
giving it all credit for its good points, and calling attention to some of its
ridiculous features. For this I got myself pretty roughly handled. The
newspapers worked themselves up to a frenzy of passion, and
committees were appointed to procure names to a petition to President
Hayes demanding my removal. The tide of popular feeling was so
violent, that I deemed it necessary to depart from my usual custom
when assailed, so far as to write the following explanatory letter, from
which the reader will be able to measure the extent and quality of my
offense:
“To the Editor of the Washington Evening Star:
“Sir:—You were mistaken in representing me as being off on a
lecturing tour, and, by implication, neglecting my duties as United
States Marshal of the District of Columbia. My absence from
Washington during two days was due to an invitation by the
managers to be present on the occasion of the inauguration of the
International Exhibition in Philadelphia.
“In complying with this invitation, I found myself in company
with other members of the government who went thither in
obedience to the call of patriotism and civilization. No one interest
of the Marshal’s office suffered by my temporary absence, as I had
seen to it that those upon whom the duties of the office devolved
were honest, capable, industrious, painstaking, and faithful. My
Deputy Marshal is a man every way qualified for his position, and
the citizens of Washington may rest assured that no unfaithful man
will be retained in any position under me. Of course I can have
nothing to say as to my own fitness for the position I hold. You
have a right to say what you please on that point; yet I think it
would be only fair and generous to wait for some dereliction of
duty on my part before I shall be adjudged as incompetent to fill
the place.
“You will allow me to say also that the attacks upon me on
account of the remarks alleged to have been made by me in
Baltimore, strike me as both malicious and silly. Washington is a
great city, not a village nor a hamlet, but the capital of a great
nation, and the manners and habits of its various classes are
proper subjects for presentation and criticism, and I very much
mistake if this great city can be thrown into a tempest of passion
by any humorous reflections I may take the liberty to utter. The city
is too great to be small, and I think it will laugh at the ridiculous
attempt to rouse it to a point of furious hostility to me for any thing
said in my Baltimore lecture.
“Had the reporters of that lecture been as careful to note what
I said in praise of Washington as what I said, if you please, in
disparagement of it, it would have been impossible to awaken any
feeling against me in this community for what I said. It is the
easiest thing in the world, as all editors know, to pervert the
meaning and give a one-sided impression of a whole speech by
simply giving isolated passages from the speech itself, without any
qualifying connections. It would hardly be imagined from anything
that has appeared here that I had said one word in that lecture in
honor of Washington, and yet the lecture itself, as a whole, was
decidedly in the interest of the national capital. I am not such a fool
as to decry a city in which I have invested my money and made
my permanent residence.
“After speaking of the power of the sentiment of patriotism I
held this language: ‘In the spirit of this noble sentiment I would
have the American people view the national capital. It is our
national center. It belongs to us; and whether it is mean or
majestic, whether arrayed in glory or covered with shame, we
cannot but share its character and its destiny. In the remotest
section of the republic, in the most distant parts of the globe, amid
the splendors of Europe or the wilds of Africa, we are still held and
firmly bound to this common center. Under the shadow of Bunker
Hill monument, in the peerless eloquence of his diction, I once
heard the great Daniel Webster give welcome to all American
citizens, assuring them that wherever else they might be
strangers, they were all at home there. The same boundless
welcome is given to all American citizens by Washington.
Elsewhere we may belong to individual States, but here we belong
to the whole United States. Elsewhere we may belong to a
section, but here we belong to a whole country, and the whole
country belongs to us. It is national territory, and the one place
where no American is an intruder or a carpet-bagger. The new
comer is not less at home than the old resident. Under its lofty
domes and stately pillars, as under the broad blue sky, all races
and colors of men stand upon a footing of common equality.
“‘The wealth and magnificence which elsewhere might oppress
the humble citizen has an opposite effect here. They are felt to be
a part of himself and serve to ennoble him in his own eyes. He is
an owner of the marble grandeur which he beholds about him,—as
much so as any of the forty millions of this great nation. Once in
his life every American who can should visit Washington: not as
the Mahometan to Mecca; not as the Catholic to Rome; not as the
Hebrew to Jerusalem, nor as the Chinaman to the Flowery
kingdom, but in the spirit of enlightened patriotism, knowing the
value of free institutions and how to perpetuate and maintain them.
“‘Washington should be contemplated not merely as an
assemblage of fine buildings; not merely as the chosen resort of
the wealth and fashion of the country; not merely as the honored
place where the statesmen of the nation assemble to shape the
policy and frame the laws; not merely as the point at which we are
most visibly touched by the outside world, and where the
diplomatic skill and talent of the old continent meet and match
themselves against those of the new, but as the national flag itself
—a glorious symbol of civil and religious liberty, leading the world
in the race of social science, civilization, and renown.’
“My lecture in Baltimore required more than an hour and a half
for its delivery, and every intelligent reader will see the difficulty of
doing justice to such a speech when it is abbreviated and
compressed into a half or three-quarters of a column. Such
abbreviation and condensation has been resorted to in this
instance. A few stray sentences, called out from their connections,
would be deprived of much of their harshness if presented in the
form and connection in which they were uttered; but I am taking up
too much space, and will close with the last paragraph of the
lecture, as delivered in Baltimore. ‘No city in the broad world has a
higher or more beneficent mission. Among all the great capitals of
the world it is preëminently the capital of free institutions. Its fall
would be a blow to freedom and progress throughout the world.
Let it stand then where it does now stand—where the father of his
country planted it, and where it has stood for more than half a
century; no longer sandwiched between two slave States; no
longer a contradiction to human progress; no longer the hot-bed of
slavery and the slave trade; no longer the home of the duelist, the
gambler, and the assassin; no longer the frantic partisan of one
section of the country against the other; no longer anchored to a
dark and semi-barbarous past, but a redeemed city, beautiful to
the eye and attractive to the heart, a bond of perpetual union, an
angel of peace on earth and good will to men, a common ground
upon which Americans of all races and colors, all sections, North
and South, may meet and shake hands, not over a chasm of
blood, but over a free, united, and progressive republic.’”
I have already alluded to the fact that much of the opposition to my

appointment to the office of United States Marshal of the District of
Columbia was due to the possibility of my being called to attend
President Hayes at the Executive Mansion upon state occasions, and
having the honor to introduce the guests on such occasions. I now
wish to refer to reproaches liberally showered upon me for holding the
office of Marshal while denied this distinguished honor, and to show
that the complaint against me at this point is not a well founded
complaint.
1st. Because the office of United States Marshal is distinct and
separate and complete in itself, and must be accepted or refused upon
its own merits. If, when offered to any person, its duties are such as he
can properly fulfill, he may very properly accept it; or, if otherwise, he
may as properly refuse it.
2d. Because the duties of the office are clearly and strictly defined
in the law by which it was created; and because nowhere among these
duties is there any mention or intimation that the Marshal may or shall
attend upon the President of the United States at the Executive
Mansion on state occasions.
3d. Because the choice as to who shall have the honor and
privilege of such attendance upon the President belongs exclusively
and reasonably to the President himself, and that therefore no one,
however distinguished, or in whatever office, has any just cause to
complain of the exercise by the President of this right of choice, or
because he is not himself chosen.
In view of these propositions, which I hold to be indisputable, I
should have presented to the country a most foolish and ridiculous
figure had I, as absurdly counseled by some of my colored friends,
resigned the office of Marshal of the District of Columbia, because
President Rutherford B. Hayes, for reasons that must have been
satisfactory to his judgment, preferred some person other than myself
to attend upon him at the Executive Mansion and perform the
ceremony of introduction on state occasions. But it was said, that this
statement did not cover the whole ground; that it was customary for the
United States Marshal of the District of Columbia to perform this social
office; and that the usage had come to have almost the force of law. I
met this at the time, and I meet it now by denying the binding force of
this custom. No former President has any right or power to make his
example the rule for his successor. The custom of inviting the Marshal
to do this duty was made by a President, and could be as properly
unmade by a President. Besides, the usage is altogether a modern
one, and had its origin in peculiar circumstances, and was justified by
those circumstances. It was introduced in time of war by President
Lincoln when he made his old law partner and intimate acquaintance
Marshal of the District, and was continued by Gen. Grant when he
appointed a relative of his, Gen. Sharp, to the same office. But again it
was said that President Hayes only departed from this custom
because the Marshal in my case was a colored man. The answer I
made to this, and now make to it, is, that it is a gratuitous assumption
and entirely begs the question. It may or may not be true that my
complexion was the cause of this departure, but no man has any right
to assume that position in advance of a plain declaration to that effect
by President Hayes himself. Never have I heard from him any such
declaration or intimation. In so far as my intercourse with him is
concerned, I can say that I at no time discovered in him a feeling of
aversion to me on account of my complexion, or on any other account,
and, unless I am greatly deceived, I was ever a welcome visitor at the
Executive Mansion on state occasions and all others, while Rutherford
B. Hayes was President of the United States. I have further to say that
I have many times during his administration had the honor to introduce
distinguished strangers to him, both of native and foreign birth, and
never had reason to feel myself slighted by himself or his amiable wife;
and I think he would be a very unreasonable man who could desire for
himself, or for any other, a larger measure of respect and consideration
than this at the hands of a man and woman occupying the exalted
positions of Mr. and Mrs. Hayes.
I should not do entire justice to the Honorable ex-President if I did
not bear additional testimony to his noble and generous spirit. When all
Washington was in an uproar, and a wild clamor rent the air for my
removal from the office of Marshal on account of the lecture delivered
by me in Baltimore, when petitions were flowing in upon him
demanding my degradation, he nobly rebuked the mad spirit of
persecution by openly declaring his purpose to retain me in my place.
One other word. During the tumult raised against me in
consequence of this lecture on the “National Capital,” Mr. Columbus
Alexander, one of the old and wealthy citizens of Washington, who was
on my bond for twenty thousand dollars, was repeatedly besought to
withdraw his name, and thus leave me disqualified; but like the
President, both he and my other bondsman, Mr. George Hill, Jr., were
steadfast and immovable. I was not surprised that Mr. Hill stood
bravely by me, for he was a Republican; but I was surprised and
gratified that Mr. Alexander, a Democrat, and, I believe, once a
slaveholder, had not only the courage, but the magnanimity to give me
fair play in this fight. What I have said of these gentlemen, can be
extended to very few others in this community, during that period of
excitement, among either the white or colored citizens, for, with the
exception of Dr. Charles B. Purvis, no colored man in the city uttered
one public word in defence or extenuation of me or of my Baltimore
speech.
This violent hostility kindled against me was singularly evanescent.
It came like a whirlwind, and like a whirlwind departed. I soon saw
nothing of it, either in the courts among the lawyers, or on the streets
among the people; for it was discovered that there was really in my
speech at Baltimore nothing which made me “worthy of stripes or of
bonds.”
Marshal at the Inauguration of Pres.
Garfield.
I can say from my experience in the office of United States Marshal
of the District of Columbia, it was every way agreeable. When it was
an open question whether I should take the office or not, it was
apprehended and predicted if I should accept it in face of the
opposition of the lawyers and judges of the courts, I should be
subjected to numberless suits for damages, and so vexed and worried
that the office would be rendered valueless to me; that it would not
only eat up my salary, but possibly endanger what little I might have
laid up for a rainy day. I have now to report that this apprehension was
in no sense realized. What might have happened had the members of
the District bar been half as malicious and spiteful as they had been
industriously represented as being, or if I had not secured as my
assistant a man so capable, industrious, vigilant, and careful as Mr.
L. P. Williams, of course I cannot know. But I am bound to praise the
bridge that carries me safely over it. I think it will ever stand as a
witness to my fitness for the position of Marshal, that I had the wisdom
to select for my assistant a gentleman so well instructed and
competent. I also take pleasure in bearing testimony to the generosity
of Mr. Phillips, the assistant Marshal who preceded Mr. Williams in that
office, in giving the new assistant valuable information as to the
various duties he would be called upon to perform. I have further to
say of my experience in the Marshal’s office, that while I have reason
to know that the eminent Chief Justice of the District of Columbia and
some of his associates were not well pleased with my appointment, I
was always treated by them, as well as by the chief clerk of the courts,
Hon. J. R. Meigs, and the subordinates of the latter (with a single
exception), with the respect and consideration due to my office. Among
the eminent lawyers of the District I believe I had many friends, and
there were those of them to whom I could always go with confidence in
an emergency for sound advice and direction, and this fact, after all the
hostility felt in consequence of my appointment, and revived by my
speech at Baltimore, is another proof of the vincibility of all feeling
arising out of popular prejudices.
In all my forty years of thought and labor to promote the freedom
and welfare of my race, I never found myself more widely and painfully
at variance with leading colored men of the country, than when I
opposed the effort to set in motion a wholesale exodus of colored
people of the South to the Northern States; and yet I never took a
position in which I felt myself better fortified by reason and necessity. It
was said of me, that I had deserted to the old master class, and that I
was a traitor to my race; that I had run away from slavery myself, and
yet I was opposing others in doing the same. When my opponents
condescended to argue, they took the ground that the colored people
of the South needed to be brought into contact with the freedom and
civilization of the North; that no emancipated and persecuted people
ever had or ever could rise in the presence of the people by whom
they had been enslaved, and that the true remedy for the ills which the
freedmen were suffering, was to initiate the Israelitish departure from
our modern Egypt to a land abounding, if not in “milk and honey,”
certainly in pork and hominy.
Influenced, no doubt, by the dazzling prospects held out to them
by the advocates of the exodus movement, thousands of poor, hungry,
naked, and destitute colored people were induced to quit the South
amid the frosts and snows of a dreadful winter in search of a better
country. I regret to say there was something sinister in this so-called
exodus, for it transpired that some of the agents most active in
promoting it had an understanding with certain railroad companies, by
which they were to receive one dollar per head upon all such
passengers. Thousands of these poor people, traveling only so far as
they had money to bear their expenses, were dropped on the levees of
St. Louis, in the extremest destitution; and their tales of woe were such
as to move a heart much less sensitive to human suffering than mine.
But while I felt for these poor deluded people, and did what I could to
put a stop to their ill-advised and ill-arranged stampede, I also did what
I could to assist such of them as were within my reach, who were on
their way to this land of promise. Hundreds of these people came to
Washington, and at one time there were from two to three hundred
lodged here, unable to get further for the want of money. I lost no time
in appealing to my friends for the means of assisting them.
Conspicuous among these friends was Mrs. Elizabeth Thompson of
New York city—the lady who, several years ago, made the nation a
present of Carpenter’s great historical picture of the “Signing of the
Emancipation Proclamation,” and who has expended large sums of her
money in investigating the causes of yellow-fever, and in endeavors to
discover means for preventing its ravages in New Orleans and
elsewhere. I found Mrs. Thompson consistently alive to the claims of
humanity in this, as in other instances, for she sent me, without delay,
a draft for two hundred and fifty dollars, and in doing so expressed the
wish that I would promptly inform her of any other opportunity of doing
good. How little justice was done me by those who accused me of
indifference to the welfare of the colored people of the South on
account of my opposition to the so-called exodus will be seen by the
following extracts from a paper on that subject laid before the Social
Science Congress at Saratoga, when that question was before the
country:
* * * * *
“Important as manual labor is everywhere, it is nowhere more

important and absolutely indispensable to the existence of society
than in the more southern of the United States. Machinery may
continue to do, as it has done, much of the work of the North, but
the work of the South requires bone, sinew, and muscle of the
strongest and most enduring kind for its performance. Labor in that
section must know no pause. Her soil is pregnant and prolific with
life and energy. All the forces of nature within her borders are
wonderfully vigorous, persistent, and active. Aided by an almost
perpetual summer abundantly supplied with heat and moisture,
her soil readily and rapidly covers itself with noxious weeds, dense
forests, and impenetrable jungles. Only a few years of non-tillage
would be needed to give the sunny and fruitful South to the bats
and owls of a desolate wilderness. From this condition, shocking
for a southern man to contemplate, it is now seen that nothing less
powerful than the naked iron arm of the negro, can save her. For
him as a Southern laborer, there is no competitor or substitute.
The thought of filling his place by any other variety of the human
family, will be found delusive and utterly impracticable. Neither
Chinaman, German, Norwegian, nor Swede can drive him from
the sugar and cotton fields of Louisiana and Mississippi. They
would certainly perish in the black bottoms of these states if they
could be induced, which they cannot, to try the experiment.
“Nature itself, in those States, comes to the rescue of the
negro, fights his battles, and enables him to exact conditions from
those who would unfairly treat and oppress him. Besides being
dependent upon the roughest and flintiest kind of labor, the climate
of the South makes such labor uninviting and harshly repulsive to
the white man. He dreads it, shrinks from it, and refuses it. He
shuns the burning sun of the fields and seeks the shade of the
verandas. On the contrary, the negro walks, labors, and sleeps in
the sunlight unharmed. The standing apology for slavery was
based upon a knowledge of this fact. It was said that the world
must have cotton and sugar, and that only the negro could supply
this want; and that he could be induced to do it only under the
“beneficent whip” of some bloodthirsty Legree. The last part of this
argument has been happily disproved by the large crops of these
productions since Emancipation; but the first part of it stands firm,
unassailed and unassailable.
“Even if climate and other natural causes did not protect the
negro from all competition in the labor-market of the South,
inevitable social causes would probably effect the same result.
The slave system of that section has left behind it, as in the nature
of the case it must, manners, customs, and conditions to which
free white laboring men will be in no haste to submit themselves
and their families. They do not emigrate from the free North, where
labor is respected, to a lately enslaved South, where labor has
been whipped, chained, and degraded for centuries. Naturally
enough such emigration follows the lines of latitude in which they
who compose it were born. Not from South to North, but from East
to West ‘the Star of Empire takes its way.’
“Hence it is seen that the dependence of the planters, land-
owners, and old master-class of the South upon the negro,
however galling and humiliating to Southern pride and power, is
nearly complete and perfect. There is only one mode of escape for
them, and that mode they will certainly not adopt. It is to take off
their own coats, cease to whittle sticks and talk politics at cross-
roads, and go themselves to work in their broad and sunny fields
of cotton and sugar. An invitation to do this is about as harsh and
distasteful to all their inclinations as would be an invitation to step
down into their graves. With the negro, all this is different. Neither
natural, artificial, or traditional causes stand in the way of the
freedman to labor in the South. Neither the heat nor the fever-
demon which lurks in her tangled and oozy swamps affright him,
and he stands to-day the admitted author of whatever prosperity,
beauty, and civilization are now possessed by the South, and the
admitted arbiter of her destiny.
“This then, is the high vantage ground of the negro; he has
labor; the South wants it, and must have it or perish. Since he is
free he can now give it or withhold it, use it where he is, or take it
elsewhere as he pleases. His labor made him a slave, and his
labor can, if he will, make him free, comfortable, and independent.
It is more to him than fire, swords, ballot-boxes, or bayonets. It
touches the heart of the South through its pocket. This power
served him well years ago, when in the bitterest extremity of
destitution. But for it, he would have perished when he dropped
out of slavery. It saved him then, and it will save him again.
Emancipation came to him, surrounded by extremely unfriendly
circumstances. It was not the choice or consent of the people
among whom he lived, but against their will, and a death struggle
on their part to prevent it. His chains were broken in the tempest
and whirlwind of civil war. Without food, without shelter, without
land, without money, and without friends, he with his children, his
sick, his aged and helpless ones, were turned loose and naked to
the open sky. The announcement of his freedom was instantly
followed by an order from his master to quit his old quarters, and
to seek bread thereafter from the hands of those who had given
him his freedom. A desperate extremity was thus forced upon him
at the outset of his freedom, and the world watched with humane
anxiety, to see what would become of him. His peril was imminent.
Starvation and death stared him in the face and marked him for
their victim.
“It will not soon be forgotten that at the close of a five hours’
speech by the late Senator Sumner, in which he advocated with
unequaled learning and eloquence the enfranchisement of the
freedmen, the best argument with which he was met in the
Senate, was that legislation at that point would be utterly
superfluous; that the negro was rapidly dying out, and must
inevitably and speedily disappear and become extinct.
“Inhuman and shocking as was this consignment of millions of
human beings to extinction, the extremity of the negro, at that
date, did not contradict, but favored, the prophecy. The policy of
the old master-class dictated by passion, pride, and revenge, was
then to make the freedom of the negro, a greater calamity to him,

Download Computational Sciences And Artificial Intelligence In Industry New Digital Technologies For Solving Future Societal And Economical Challenges And Automation Science And Engineering 76 Tero Tuo online ebook texxtbook full chapter pdf

Uploaded by

Copyright:

Available Formats

You might also like

Download Computational Sciences And Artificial Intelligence In Industry New Digital Technologies For Solving Future Societal And Economical Challenges And Automation Science And Engineering 76 Tero Tuo online ebook texxtbook full chapter pdf

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Download Computational Sciences And Artificial Intelligence In Industry New Digital Technologies For Solving Future Societal And Economical Challenges And Automation Science And Engineering 76 Tero Tuo online ebook texxtbook full chapter pdf

Uploaded by

Copyright:

Available Formats

Computational Sciences and Artificial

Intelligence in Industry: New Digital

Artificial Intelligence, Machine Learning, and Data

Automation and Autonomy: Labour, Capital and Machines

Semantic Web Technologies Research and Applications

Primary Mathematics 3A Hoerst

Artificial Intelligence: Technologies, Applications,

Computational Intelligence and Data Sciences: Paradigms

Software Engineering Artificial Intelligence Networking

Artificial Intelligence and Machine Learning in Public

Panos Antsaklis, University of Notre Dame, USA

More information about this series at http://www.springer.com/series/6259

ISSN 2213-8986 ISSN 2213-8994 (electronic)

The CSAI 2019 Conference (Computational Science and AI in Industry: New

This chapter is focused on the development of computational science, artiﬁcial

derivative technologies are presented, with some challenges related to scientiﬁc

The main content of Part II includes:

Part III Medical Applications

The main content of Part III includes:

The main content of Part IV includes:

Perspectives in Industry and Society:

In conclusion to this preface, let us quote a say of the internationally renowned

Jyväskylä, Finland Tero Tuovinen

Improving Clustering and Cluster Validation with Missing Data

Part III Medical Applications

Part IV Industrial and Technological Applications

Esa Alakoski JAMK University of Applied Science, Jyväskylä, Finland

Cameron James Adagos, Ramonville-Saint-Agne, France;

Olivier Pironneau LJLL, Sorbonne Université, Paris, France

Pekka Neittaanmäki, Matti Savonen, Jacques Periaux, and Tero Tuovinen

Abstract In this introductory chapter of the book, we outline the development of

1 Simultaneous Development of Hardware and Software

Simultaneous development of hardware and software is critically important. Devel-

© Springer Nature Switzerland AG 2022 3

Fig. 1 Evolution of various scientific paradigms: hardware development accelerating scientific

result = algorithm × hardware.

Fig. 2 Research progress in telecommunications: advances in physics-related disciplines (top line)

2 Co-development Potential of Computational Science and

As a result of deep learning advances, natural language processing solutions, like

Fig. 3 Artificial intelligence terminology and derivative technologies

3 Challenges in the Contemporary Field of Scientific

Swagata Roy and Nirupam Chakraborti

S. Roy · N. Chakraborti (B)

© Springer Nature Switzerland AG 2022 11

where X = {xk | k = 1, 2, . . . , K } is a vector of variables, and i = { f i | i =

(i ≺ m ) ↔ (∀i )( f il ≤ f im ) ∧ (∃m)( f il < f i m) (3)

Fig. 1 A KIMEME interface with BioGP incorporated

Fig. 2 A typical EvoDN2 architecture

Algorithm 2 Pseudocode of EvoDN2

Reference Vector guided Evolutionary Algorithm (RVEA) [3] is a multi-objective

as constraint-based Reference Vector Evolutionary Algorithm (cRVEA) [3], which

Fig. 3 Training of ZDT3 second objective by EvoDN2

4 Applications in Materials Research

The genesis of metamodeling algorithms BioGP, EvoNN, and EvoDN2 is discussed

Pekka Neittaanmäki and Sergey Repin

Keywords Scientific computations · Machine learning · Artificial intelligence ·

1 Classical Method of Scientific Knowledge

P. Neittaanmäki (B) · S. Repin

Scientific computing is a multidisciplinary field that uses advanced computing