Professional Documents
Culture Documents
PDF Artificial Intelligence Applications and Innovations 15Th Ifip WG 12 5 International Conference Aiai 2019 Hersonissos Crete Greece May 24 26 2019 Proceedings John Macintyre Ebook Full Chapter
PDF Artificial Intelligence Applications and Innovations 15Th Ifip WG 12 5 International Conference Aiai 2019 Hersonissos Crete Greece May 24 26 2019 Proceedings John Macintyre Ebook Full Chapter
https://textbookfull.com/product/engineering-applications-of-
neural-networks-20th-international-conference-
eann-2019-xersonisos-crete-greece-may-24-26-2019-proceedings-
john-macintyre/
https://textbookfull.com/product/cloud-computing-and-services-
science-9th-international-conference-closer-2019-heraklion-crete-
greece-may-2-4-2019-revised-selected-papers-donald-ferguson/
Artificial Intelligence for Communications and
Networks: First EAI International Conference, AICON
2019, Harbin, China, May 25–26, 2019, Proceedings, Part
I Shuai Han
https://textbookfull.com/product/artificial-intelligence-for-
communications-and-networks-first-eai-international-conference-
aicon-2019-harbin-china-may-25-26-2019-proceedings-part-i-shuai-
han/
https://textbookfull.com/product/artificial-intelligence-
xxxvi-39th-sgai-international-conference-on-artificial-
intelligence-ai-2019-cambridge-uk-
december-17-19-2019-proceedings-max-bramer/
https://textbookfull.com/product/testing-software-and-
systems-31st-ifip-wg-6-1-international-conference-
ictss-2019-paris-france-october-15-17-2019-proceedings-
christophe-gaston/
https://textbookfull.com/product/trust-management-xiii-13th-ifip-
wg-11-11-international-conference-ifiptm-2019-copenhagen-denmark-
july-17-19-2019-proceedings-weizhi-meng/
IFIP AICT 559
John MacIntyre
Ilias Maglogiannis
Lazaros Iliadis
Elias Pimenidis
(Eds.)
Artificial Intelligence
Applications
and Innovations
123
IFIP Advances in Information
and Communication Technology 559
Editor-in-Chief
IFIP is the global non-profit federation of societies of ICT professionals that aims
at achieving a worldwide professional and socially responsible development and
application of information and communication technologies.
Artificial Intelligence
Applications
and Innovations
15th IFIP WG 12.5 International Conference, AIAI 2019
Hersonissos, Crete, Greece, May 24–26, 2019
Proceedings
123
Editors
John MacIntyre Ilias Maglogiannis
University of Sunderland University of Piraeus
Sunderland, UK Piraeus, Greece
Lazaros Iliadis Elias Pimenidis
Democritus University of Thrace University of West England
Xanthi, Greece Bristol, UK
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
AIAI 2019
According to Professor Klaus Schwab (founder and executive chairman of the World
Economic Forum), we are living in the era of a great revolution that is rapidly bringing
huge changes and challenges in our daily lives. This is the Fourth Industrial Revolu-
tion, which has a big impact on all disciplines, even in the way that we communicate
and interact with each other. Artificial intelligence (AI) is a major and significant part
of the Fourth Industrial Revolution. Its rapid technical breakthroughs are enabling
superhuman performance by machines in a real-time mode. Machine vision (e.g., face
recognition) or language translators and assistants like Siri and Alexa are characteristic
examples. AI is promising a brave new world where business and economies will
expand their productivity and innovation. Machine learning and deep learning are part
of our usual common interactions on our mobile phones and on social media.
Numerous applications of AI are used in almost all domains from cybersecurity to
financial and medical cases. However, historic challenges for the future of mankind are
being faced. Potential unethical use of AI may violate democratic human rights and
may alter the character of Western societies.
The 15th Artificial Intelligence Applications and Innovations (AIAI) conference
offered insight into all timely challenges related to technical, legal, and ethical aspects
of intelligent systems and their applications. New algorithms and potential prototypes
employed in diverse domains were introduced.
AIAI is a mature international scientific conference held in Europe and is well
established in the scientific area of AI. Its history is long and very successful, following
and spreading the evolution of intelligent systems.
The first event was organized in Toulouse France in 2004. Since then, it has had a
continuous and dynamic presence as a major global, but mainly European scientific
event. More specifically, it has been organized in China, Greece, Cyprus, Australia, and
France. It has always been technically supported by the International Federation for
Information Processing (IFIP) and more specifically by the Working Group 12.5, which
is interested in AI applications.
Following a long-standing tradition, this Springer volume belongs to the IFIP AICT
Springer Series and it contains the papers that were accepted to be presented orally at
the AIAI 2019 conference. An additional volume comprises the papers that were
accepted and presented at the workshops and were held as parallel events.
vi Preface
Recommender systems
Intelligent telecommunications modeling
Intelligent hybrid systems using internet of things
The authors of submitted papers came from 23 different countries from all over the
globe, namely: Austria, Brazil, Canada, Colombia, Czech Republic, Finland, France,
Germany, Greece, The Netherlands, Hungary, India, Italy, Japan, P.R. China, Peru,
Poland, Portugal, Spain, Sweden, Tunisia, the UK, and the USA.
Executive Committee
General Chairs
John MacIntyre University of Sunderland, UK (Dean of the Faculty
of Applied Sciences and Pro Vice Chancellor
of the University of Sunderland)
Ilias Maglogiannis University of Piraeus, Greece
(President of the
IFIP WG12.5)
Plamen Angelov University of Lancaster, UK
Program Chairs
Lazaros Iliadis Democritus University of Thrace, Greece
Elias Pimenidis University of the West of England, Bristol, UK
Advisory Chairs
Stefanos Kolias University of Lincoln, UK
Spyros Likothanasis University of Patras, Greece
Georgios Vouros University of Piraeus, Greece
Honorary Chair
Barbara Hammer Bielefeld University, Germany
Workshop Chairs
Christos Makris University of Patras, Greece
Phivos Mylonas Ionian University, Greece
Spyros Sioutas University of Patras, Greece
Program Committee
Michel Aldanondo IMT Mines Albi, France
Athanasios Alexiou NGCEF, Australia
Mohammed Alghwell Freelancer, Libya
Ioannis Anagnostopoulos University of Central Greece, Greece
George Anastassopoulos Democritus University of Thrace, Greece
Vardis-Dimitris Anezakis Democritus University of Thrace, Greece
x Organization
Panagiotis Papapetrou
Abstract. The first part of the talk will focus on data mining methods for
learning from Electronic Health Records (EHRs), which are typically perceived
as big and complex patient data sources. On them, scientists strive to perform
predictions on patients’ progress, to understand and predict response to therapy,
to detect adverse drug effects, and many other learning tasks. Medical
researchers are also interested in learning from cohorts of population-based
studies and of experiments. Learning tasks include the identification of disease
predictors that can lead to new diagnostic tests and the acquisition of insights on
interventions. The talk will elaborate on data sources, methods, and case studies
in medical mining.
The second part of the talk will tackle the issue of interpretability and
explainability of opaque machine learning models, with focus on time series
classification. Time series classification has received great attention over the past
decade with a wide range of methods focusing on predictive performance by
exploiting various types of temporal features. Nonetheless, little emphasis has
been placed on interpretability and explainability. This talk will formulate the
novel problem of explainable time series tweaking, where, given a time series
and an opaque classifier that provides a particular classification decision for the
time series, the objective is to find the minimum number of changes to be
performed to the given time series so that the classifier changes its decision to
another class. Moreover, it will be shown that the problem is NP-hard. Two
instantiations of the problem will be presented. The classifier under investigation
will be the random shapelet forest classifier. Moreover, two algorithmic solu-
tions for the two problem instantiations will be presented along with simple
optimizations, as well as a baseline solution using the nearest neighbor classifier.
Empirical Approach: How to Get Fast,
Interpretable Deep Learning
Plamen Angelov
Evangelos Eleftheriou
Invited Paper
Biomedical AI
Classification - Clustering
Fuzzy Modeling
Recommendation Systems
1 Introduction
The second author gratefully acknowledges the partial support of NSERC, the Natural
Sciences and Engineering Council of Canada.
c IFIP International Federation for Information Processing 2019
Published by Springer Nature Switzerland AG 2019
J. MacIntyre et al. (Eds.): AIAI 2019, IFIP AICT 559, pp. 3–16, 2019.
https://doi.org/10.1007/978-3-030-19823-7_1
4 A. Shirvani and B. John Oommen
used to update the action probabilities, instead of updating it based only on the
RE’s feedback. In this way, the probabilities of choosing the actions with higher
reward estimates were increased, and those with lower estimates were signifi-
cantly reduced, using which they proposed the family of estimator algorithms.
The Pursuit strategy of designing LA is a special derivative of the family
of estimator algorithms. Pursuit algorithms “pursue” the currently-known best
action, and increase the action probability associated with this action. The pur-
suit concept was first introduced by Thathachar et al., and the corresponding
LA was proven to be -optimal. Its discretized version was proposed by Lanc-
tot et al. in [7], who also discretized the original estimator algorithm. Agache
et al. [10] then analyzed all the four linear combinations, i.e., the LRI and LRP
paradigms.
The Object Partitioning Problem (OPP): Consider the problem of par-
titioning a set A = {A1 , · · · , AW } of W physical objects into R groups
Ω = {G1 , · · · , GR }. We assume that the true but unknown state of nature,
Ω ∗ , is a partitioning of the set A into mutually exclusive and exhaustive sub-
sets {G∗1 , G∗2 , · · · G∗R }. The composition of {G∗i } is unknown, and the elements in
the subsets fall together based on some criteria which may be mathematically
formulated, or may even be ambiguous. These objects are now presented to a
learning algorithm, for example, in pairs or tuples. The goal of the algorithm is
to partition A into a learned partition, Ω + . The hope is to have Ω + converge
to Ω ∗ . In most cases, the underlying partitioning of Ω ∗ is not known, nor are
the joint access probabilities by which the pairs/tuples of A are presented to the
learning algorithm known. This problem is known to be NP-hard [9]. Clearly,
if we increase the number of objects, the number of partitions increases, and
in addition to this quantity, the problem’s complexity grows exponentially. To
resolve this, it is possible to explore all partition combinations, use a ranking
index, and to thereafter, report the best plausible partition. The goal of the OPP
is to identify the best or most likely realizable partitioning. This requires the AI
algorithm to perceive the semantic physical world aspects of the objects, and
to then make local decisions based on the best partition in the abstract domain
[4,5].
Real versus Abstract Objects: If there exists a mutual relation between
the real objects in the semantic domain A, and a domain of abstract objects
O = {O1 , · · · , OW }, we define the partitioning of O in such way that the cor-
responding partitions of O map onto the partitions of the real objects in A so
as to mimic the state-of-nature. Thus, while we operate on the abstract objects
in O, the objects in A are not necessarily moved because they constitute real-
life objects which cannot be easily moved. A special case of the OPP is the
Equi-Partitioning Problem (EPP) in which all the partitions are equi-sized.
The Object Migrating Automation (OMA): Due to the poor convergence
of prior OPP/EPP solutions, they were never utilized in real-life applications.
The introduction of an LA-based partitioning algorithm, the OMA, (explained
in Sect. 2) made real-life applications possible. The OMA resolved the EPP both
The Power of the Pursuit Learning Paradigm in the Partitioning of Data 5
efficiently and accurately. This solution is regarded as a benchmark for the EPP.
Indeed, since 19861 , it has been applied to variety of real-life problems and
domains which include keyboard optimization, image retrieval, distributed com-
puting, graph partitioning, the constraint satisfaction problems, cryptanalysis,
reputation systems, parallel and distributed mapping etc.
The Intent of this Paper: Although the “Pursuit” learning paradigm has
been utilized in the theory and applications of LA as fundamental machines,
until recently, it has not been incorporated into solutions that solve partitioning
problems. The goal of this paper is to submit a comprehensive survey of how this
paradigm can be used in Object Partitioning, and to optimize various versions of
the OMA. We also include simulation results on benchmark environments that
demonstrate the advantages of incorporating it into the respective machines.
1
The bibliography in this paper is necessarily limited. The majority of the present
results very briefly summarize the results in the Ph.D. thesis of the First Author.
2
To be consistent with the terminology of LA, we use the terms “action”, “class” and
“group” synonymously.
3
The OMA’s algorithms/figures are in [11], and omitted here in the interest of space.
6 A. Shirvani and B. John Oommen
2. If ξi = φkN or ξj = φmN but not both (i.e., only one of these abstract
objects is in the boundry state), the object which is not in the bound-
ary state, say Oi , is moved towards its boundary state as shown in
Figure 3.2(c) in [11]. Simultaneously, the object that is in the boundary
state, Oj , is moved to the boundary state of Oj . Since this reallocation
will result in an excess of objects in αk , we choose one of the objects in
αk (which is not accessed) and move it to the boundary state of αm . In
this case, we choose the object nearest to the boundary state of ξi , as
shown in Figure 3.2(c) in [11].
3. If ξi = φkN and ξj = φmN (both objects are in the boundary states),
one object, say Oi , will be moved to the boundary state of αm . Since this
reallocation, will again, result in an excess of objects in αm , we choose one
of the objects in αm (which is not accessed) and move it to the boundary
state of αk . In this case, we choose the object nearest to the boundary
state of ξj , as shown in Figure 3.2(d) in [11].
To asses the partitioning accuracy and the convergence speed of any EPP
solution, there must be an “oracle” with a pre-defined number of classes, and with
each class containing an equal number of objects. The OMA’s goal is to migrate
the objects between its classes, using the incoming queries. E is characterized by
three parameters: (a) W , the number of objects, (b) R, the number of partitions,
and (c) a probability ‘p’ quantifying how E pairs the elements in the query.
Table 1. Experimental results for the OMA done for an ensemble of 100 experiments
in which we have only included the results from experiments where convergence has
occurred.
where P (Rk ) is the probability that the first element, Ai , is chosen from the
group Rk , and P (Aj |Ai ) is the conditional probability of choosing Aj , which
is also from Rk , after Ai has been chosen. Since E chooses the elements of the
pairs from the other groups uniformly, with a possible re-numbering operation,
the matrix M∗ = [μ∗ (i, j)] is a block-diagonal matrix given by Eq. (1).
⎡M∗ 0 . . . 0 ⎤
1
⎢ .. ⎥
⎢ 0 M∗2 . ⎥
∗ ⎢
M =⎣ .. . ⎥
. .. ⎦
.. (1)
.
0 · · · . . . M∗R
⎡ ⎤
0 R
W (W −R) ··· R
W (W −R)
⎢ R
··· R ⎥
⎢ W (W −R) 0 W (W −R) ⎥
∗
Mr = ⎢ ⎢ ⎥ (2)
.. .. .. ⎥
⎣ . . . ⎦
R
W (W −R) ··· R
W (W −R) 0
Another random document with
no related content on Scribd:
water on it and measure the heat imparted to the water in a given
time, also the input to the heating element in the same time, from
which data the efficiency may be calculated. In the case of an
electric iron, dampened cloths may be ironed and the actual water
evaporated by the iron, determined by weighing the cloths before
and after the ironing, together with the increase in weight of the cloth
on the ironing board, the time the iron is in use and the temperature
of the cloths. The actual water evaporated is the difference in the
weight of the cloths before and after ironing, minus the increase in
weight of the cloth on the ironing board, which takes up some of the
moisture from the cloths being ironed.
Earthen Mustard Pots Used as Acid Jars
Fig. 2 Fig. 4
Owners of sail or power boats will find the take-down oars shown
in the sketch easily made and of value in an emergency far out of
proportion to the space occupied in a boat. A pair of ordinary oars
was cut as shown, and pipe fittings were attached to the ends to
form a detachable joint. When knocked down the oars may be stored
in a seat cupboard, or other convenient place.—Contributed by H. E.
Ward, Kent, Wash.
How to Make Propeller Blades Quickly
Requiring a number of propeller blades for use in making models
of windmills, and other constructions, I found that I could save much
time and obtain a satisfactory set of propeller blades by using
ordinary shoehorns of the same size. The small ends of the horns
were flattened out so that they could be fastened to pieces of wood
for bearings, and then hammered to the proper shape for cutting the
air, or receiving the force of the wind.
Bench Stop