“Adapting the Teaching of Computational Intelligence Techniques to Improve Learning Outcomes”

Adapting the Teaching of Computational
Intelligence Techniques to Improve
Learning Outcomes

Thomas Hanne and Rolf Dornberger

Abstract In the Master of Science program Information Systems at a
Swiss university, the authors have been teaching artificial intelligence (AI) methods,
in particularly computational intelligence (CI) methods, for about ten years. AI and
CI require the ability and readiness of a deeper understanding of algorithms, which
can hardly be achieved with classical didactic concepts. Therefore, the focus is on
assignments that lead the students to develop new algorithms or modify existing
ones, or make them suitable for new areas of applications. This article discusses
certain teaching concepts, their changes over time and experiences that have been
made with a focus on improving students’ learning outcomes in understanding and
applyingspecialAI/CImethodssuchasneuralnetworksandevolutionaryalgorithms.

Keywords Computational intelligence · Artificial intelligence · Teaching ·
Learning assessment · STEM

1 Introduction

In research as well as in the labor market, STEM skills are strongly required—beyond
all technical disciplines—also in business and society in general, and often provide
above-average job and income opportunities for qualified students [1]. However,
teaching STEM subjects (Sciences, Technology, Engineering, and Mathematics)
provides particular challenges (see, e.g. [2]). Students often have difficulties in under-
standing and learning the respective subjects, which may have other reasons than real

T. Hanne (B)
School of , Institute for Information Systems, FHNW University of Applied Sciences and
Arts Northwestern Switzerland, Riggenbachstrasse 16, 4600 Olten, Switzerland
e-mail: [email protected]

R. Dornberger
School of , Institute for Information Systems, FHNW University of Applied Sciences and
Arts Northwestern Switzerland, Peter Merian-Strasse 86, 4002 Basel, Switzerland
e-mail: [email protected]

© Springer Nature Switzerland AG 2021
R. Dornberger (ed.), New Trends in Information Systems and Technology,
Studies in Systems, Decision and Control 294,
https://doi.org/10.1007/978-3-030-48332-6_8

113

http://crossmark.crossref.org/dialog/?doi=10.1007/978-3-030-48332-6_8&domain=pdf

https://orcid.org/0000-0002-5636-1660

mailto:[email protected]

mailto:[email protected]

https://doi.org/10.1007/978-3-030-48332-6_8

114 T. Hanne and R. Dornberger

or imaginary deficiencies in their respective competencies (e.g. analytical thinking)
to insufficient pre-arrangements or inadequate teaching methods.

Computational intelligence (CI), comprising mainly the nature-inspired artificial
intelligence (AI) methods and further metaheuristics, is one of the fields of science
at the cutting edge of STEM disciplines. It has become particularly important during
the last ten years along with the new rise of artificial intelligence (see, e.g. [3]), which
is assumed to provide some of the most important changes and disruptions in society
since the introduction of computer and information technologies in the middle of the
last century.

Usually, CI is not specified by an unambiguous definition but by enumerating
subareas (see, e.g. [4]), in particular

• Fuzzy logic
• Neural networks
• Evolutionary computation
• Swarm intelligence.

Fuzzy logic provides theoretical insights, modeling approaches and methods to
cope with problems, which can hardly be described by traditional binary logic and
therefore contain elements of vagueness of knowledge in order to cope with uncer-
tainty similar to human reasoning. (Artificial) neural networks mimic the function
of real nerves and nerve nets as observed in animals and humans. It is one of the
most important approaches based on machine learning in current AI developments
and, nowadays, finds particular attention in the field of deep learning. Evolutionary
computation is based on a simulation of biological evolutionary processes (as first
described by Darwin) to find superior (optimal) solutions to complex problems. Even
with moving objectives, evolutionary computation is able to continuously adapt the
solution to new conditions just as biological evolution. Swarm intelligence comprises
a number of similar approaches to solving complex problems, which are based on
strategies found in biology, especially the behavior of animal swarms (i.e. flocks,
packs, hives), which may emerge to a complex problem solving behavior, which is
not shown in the individual behavior of animals.

The CI algorithms derived from these approaches are used to find good alternative
solutions, optimizing candidate solutions, identifying patterns in data, and mapping
input data to possible outputs. In general, CI methods work in a static context of
the problem, as well as in time-dependent, changing problems, where some of these
CI methods are also used for controller design. A recent development is to apply
CI methods in robotics to make the robots “more intelligent”, as swarm intelligence
allows the self-organizing of a bunch of robots in a swarm.

These approaches have in common that they can deal with poorly structured and/or
difficult-to-solve problems and mostly rely on solution concepts that are adapted
from processes in nature. Therefore, such concepts are often denoted nature-inspired
methods. As they are often based on incomplete or uncertain knowledge or allow for
good, although not optimal solutions, e.g. by using heuristics or metaheuristics, soft
computing or nature-inspired computation are other expressions to refer to this field
of science.

Adapting the Teaching of Computational Intelligence … 115

The competencies to understand and apply CI are manifold. On the one hand, there
are the algorithms, which require skills in programming and software engineering
for understanding and coding the algorithms or for their adaptation or further devel-
opment. On the other hand, the treated problems are often complex optimization
problems that require a significant mathematical understanding to define the opti-
mization problem with its search space and constraints correctly. Mathematics is
also (for some part) necessary to understand and analyze the respective algorithms.
In addition, the power of the CI methods lies in solving real-world problems, which
requires profound skills to abstract the complexity of the world up to computational
models, as e.g. in computational sciences or operations research.

The goal of this article is to present and discuss the authors’ special teaching
concepts. The changes of the teaching concepts over time and the authors’ expe-
riences are discussed with a focus on improving the students’ learning outcomes
in understanding and applying special AI/CI methods. In the presented case, the
focus is set on the personal experiences while teaching neural networks and evolu-
tionary algorithms and similar heuristics. However, it can be assumed that many
other lecturers teaching AI methods face similar problems and will benefit from the
reported measures and generalized statements.

In Sect. 2 of this chapter, we discuss selected related work. Our teaching concepts
related to CI are described in Sect. 3. In Sect. 4, we discuss reasons for several
changes of the course concept during the last ten years. Evaluation aspects regarding
the course success are considered in Sect. 5. Conclusions are provided in Sect. 6.

2 Related Work

In general, there is very few published knowledge about teaching computational intel-
ligence. It is possible to find various more or less detailed descriptions of university
courses related to CI, but few insights into why they were designed as they are. In
addition, little is known about the evaluation of different course designs.

Although [5] explicitly addresses CI, there is little specific insight into how to
teach such topics. In [6] more specific aspects of CI teaching are considered. As
one of the few concrete examples related to individual CI courses let us mention the
paper by [7]. The didactic setting has some similarities with our teaching approaches,
which we report on later, e.g. a student project assignment, but stronger focus is laid
on traditional teaching and assessment. For instance, the project work only makes up
30% of the overall mark, whereas it is 70% in our case. While that course is mainly
intended for undergraduate students, ours is offered at the Master level. Unlike the
course described, our study program no longer includes any more specialized follow-
up courses. Both courses have in common that they strongly support the fact that the
project results should be sufficient for a scientific publication.

In [8], an elective graduate-level course on CI related to engineering applications is
considered, which strongly focuses on term projects similar to our course. In addition
to the standard CI approaches, the use of swarm robotics is mentioned, which is a

116 T. Hanne and R. Dornberger

topic also frequently offered in our course. In addition, summer internships related
to CI are offered in the considered study program. Apart from some general remarks
on STEM education, the didactic setting of the training under consideration is not
described in detail. This also holds for evaluation results. Samanta and Turner [9]
addresses a course, which also includes some CI contents, but the main focus is
on mechatronics and robotics. Stachowicz [10] is another example that focuses on
a single CI-related course. Most of the time, however, content is described, while
details about the educational settings remain largely unclear. In [11], an overview of
CI courses at different universities is provided. In most cases, however, only content
is briefly described, while educational settings are not further discussed.

In some cases, such as [12], insights are reported from courses that focus on
more specific content (such as a design optimization method) in a related university
course. Another example is [13], which focuses on a tool based on a particular CI
method (particle swarm optimization) for its use in a teaching setting. However,
from our perspective, there are no particular reasons to use the specific “teaching
implementations”ofsuchmethods.Incontrast,weassumethatitmakesmoresenseto
use regular implementations, which might be a better choice to have more freedom to
use them later, e.g. when working in a company. Some other publications focus on the
use of gamification or serious games related to teaching computational intelligence
(e.g. [14]).

However, if we consider the much larger area of STEM disciplines, there is quite
a lot of published work related to teaching aspects. We only mention a few of them,
since they mostly give us only rather general insights of what might be a useful
setting for a CI-related course. Often, these publications are too broad to allow for
useful conclusions and practical applications. For instance, [15] discusses ideological
aspects rather than concrete problems and suggestions related to the teaching of
information and communication technologies (ICT).

The focus in [16] is also quite broad and includes different bachelor, master, and
Ph.D. study programs in the areas Financial Management, Management of Tourism,
Applied Computer Science, Information Management, and Information and Knowl-
edge Management. The derived recommendations are quite concrete and based on
abilities and skills required in industry and business. However, recommendations are
not as detailed as required for the design of individual courses.

An even broader survey of study settings in Computational Science and Engi-
neering is given in [17], which analyzes respective study programs from several
universities including the content of courses, curricula, and the degrees offered.
Although the topic appears to be more specific, the focus here is not on one single
institution, and details of teaching concepts are mainly missing. In [18], the situation
in teaching in electrical and computer engineering is considered. While the main
focus of this paper is quite broad, it provides some insights that are also relevant to
our teaching of CI, such as the emphasis on project–problem-based learning.

Detailed aspects of learning in mathematics are considered in [19]. For instance,
they point out the learner’s difficulties in the (re-)construction of general mathemat-
ical knowledge. Since more general mathematics content than in CI is considered,

Adapting the Teaching of Computational Intelligence … 117

the aspects of problem-based learning and individual assignments are not so much
taken into account.

Although teaching concepts for STEM subjects and selected AI methods (e.g.
decisionrules)atschoolanduniversitylevelarequitewellresearchedandunderstood,
guidance to teach sophisticated CI methods (such as evolutionary computing, swarm
intelligence, etc.) at university level is still lacking. That is why we present the case of
reporting, evaluating, and discussing our continuous teaching approaches and share
our experiences from more than ten years of lecturing CI at the University of Applied
Sciences and Arts Northwestern Switzerland FHNW.

3 CI in the Master of Science Program in
Information Systems

We teach CI within a course of the Master of Science (M.Sc.) program in
Information Systems (BIS) at a school of business in Switzerland. The study program
can be studied full- or part-time, which is rather an administrative distinction of an
assumed study period of three or five semesters in order to obtain the required credit
points. The M.Sc. BIS includes four core courses and 14 electives including the CI
course to provide substantial opportunities for choice and specialization among the
students. The core courses are not or only marginally related to the required skills
in programming or mathematics. The only preparation in the M.Sc. BIS program
provided for programming-related aspects is a short pre-course in programming
that will give at least a short introduction to software development, algorithms, and
programming to those students holding a bachelor degree in Administration
or similar areas, and who have little background in these matters. Consequently, there
are almost no basics from the study program, which could facilitate the learning in
the CI course.

In addition, the students come from heterogeneous backgrounds in relation to their
countries of origin and their culture, but also taking into account the knowledge they
have acquired in their previous bachelor’s studies, their apprenticeships, and work
experience. Some of them have a bachelor degree in information systems, while
others come from other areas such as business administration, computer science,
engineering, or social sciences. Especially students with a business administration
background make up a significant part of the student population with no or little
background in programming and software engineering. Moreover, the mathematical
background among the students is often rather weak, because many of them followed
the way of apprenticeships with more practice-oriented teaching and learning instead
of a stringent high school education.

Another aspect is that our university belongs to the group of “Universities of
Applied Sciences”, which put a stronger focus on practically applied content. While
this is preferred for subsequent work in business and industry, theoretical skills (such

118 T. Hanne and R. Dornberger

as in mathematics) can be taught in a reduced way. The above-mentioned difficulties
are particularly relevant for teaching the CI course.

A further aspect that impedes the acquisition of missing skills during the course
semester is the limited amount of time among students. In addition to other learning
requirements and personal requests, today, more than two thirds of the students study
part-time and have a workload from their jobs of about 60%, occasionally up to 100%
of full-time workload. Today, we observe that the focus of many students has shifted
away from the idea of learning new topics at the university, as students want to obtain
a Master’s degree with minimal effort.

The course on CI was introduced in 2009, shortly after the start of our master
program in Information Systems in 2008. In the meantime, students can
begin the M.Sc. BIS program twice a year, in autumn and spring. The course on CI is
based on approximately twelve lecturing blocks of four teaching hours (4 × 45 min)
and one presentation block during a semester of about 15 weeks [20]. The course
is usually completed by students in the second, third, or fourth semester. Classroom
lessons are used for traditional lectures, exercises, student presentations, and discus-
sions. According to the European Credit Transfer and Accumulation System (ECTS),
the course yields six credit points, which corresponds to approximately 180 working
hours per student (including self-study time).

The lecturing part is mainly for teaching basics in CI with some focus on busi-
ness applications as discussed in [21]. From the CI subareas mentioned above, we
put a stronger emphasis on the subareas of evolutionary computation and swarm
intelligence because we assume that these fields are still underdeveloped in various
practical applications (see, e.g. [22]). The more important part of the course is based
on student assignments. In these assignments, the students are expected to familiarize
themselveswithaparticularproblem(usuallyanoptimizationproblem)and/orasolu-
tion approach (usually an optimization method), usually based on a given publication.
In addition, the following types of tasks are typical for the assignments:

• The considered problem should be solved by applying a problem-solving method
different from the original problem-solving method,

• a different type of problem (similar to the considered one) has to be solved with
the original problem-solving method or another one,

• variations of the methods are explored,
• computational studies are done, which are more extensive than those reported in

the literature.

For that purpose, the students (in groups or alone) are given sample reference
papers related to their chosen topic and are expected to make themselves familiar
with the optimization problem and/or suitable methods for solving it. Then they
search for additional references, e.g. other problem types or further methods, which
could be used. In the following, the students are expected to implement another
method to solve a given problem, or to apply a given method to a different type of
problem, or to do some modifications regarding an existing method.

In the past, we have strongly recommended that the students use the OpenCI soft-
ware suite for this implementation work, which has been developed at our university

Adapting the Teaching of Computational Intelligence … 119

in cooperation with some other institutions [23, 24]. In recent years, we have opened
the choice of a suitable software platform: We also encourage the students to use
a different framework or implement or modify a stand-alone code. After that, they
are expected to complete computational tests with the considered settings regarding
problems and solution approaches (e.g. exploring different parameter settings) and
a subsequent analysis of results. In the end, they have to submit a scientific paper
that should be similar to a typical conference paper, and all the artifacts elaborated
during this project, such as the developed software, a short user guide, and the data
used.

As discussed above, these tasks present significant difficulties for the students due
to the required mathematical and software engineering background. In particular,
we try to alleviate the software engineering difficulties in the following ways: In the
beginning of the course, a general introduction to the programming-related aspects is
given, including an introduction to the Java-based OpenCI framework. For instance,
the students learn about the installation and architecture of this tool and related
software such as the integrated development environment Eclipse. In addition, they
learn about the OpenCI architecture and how to embed a new solution algorithm or
a new type of problem into OpenCI. OpenCI is meanwhile a rather large software
framework including various implementations of problem representations, solution
algorithms (basic algorithms for search and optimization and machine learning), a
graphical user interface, and further tools for visualization and data analysis, and
additional libraries. Therefore, it is already a challenge to understand the basic ideas
of this framework despite the endeavor from our side to make it as simple as possible.
The maintenance of OpenCI and the integration of the students’ implementation into
one common version of OpenCI is always a great challenge for us lecturers.

In the software development-related part within the CI project, the students learn
how to embed a new problem representation as a (Java) class and/or a new algo-
rithm, each of them being represented as classes. While the lecturing part serves a
better theoretical understanding of problems and solutions (including some related
mathematics), the students receive a basic preparation for writing a scientific paper
in a course on research methodology and a related project to be elaborated at the
beginning of the study program. However, bringing everything together—software
development, mathematics, scientific writing, and the specific contents treated in the
CI course—is certainly a challenging task.

It has turned out to be useful to define milestones for the student projects such
as intermediate presentations of their work. On the one hand, this allows us (i.e. the
lecturers, but also other students) to give feedback and discuss problems occurring
in a student group. On the other hand, the students are requested to start early with
their work and work continuously during the semester, which should help to avoid
situations with too much time pressure at the end of their projects. During previous
semesters of the course, there were one or two mandatory presentations without
grading. During the last semester of the course (spring semester, 2019) we changed
that to a voluntary presentation, but with (mandatory) graded deliverables in form
of short papers related to the topics. On the one hand, the grading aspect should put
more emphasis on the quality of these intermediate deliverables. On the other hand,

120 T. Hanne and R. Dornberger

we skipped any other deliverables for grading apart from the student assignment
(research paper) and the two intermediate papers.

Whenwestartedthemodulein2009,itwasaccompaniedbyaregularwrittenexam
that focused on the contents of the lecture part, i.e. mostly basic CI contents. Later
(in 2015), we changed this to an oral exam at the end of the semester, which included,
in addition to already taught contents, a part related to the group assignments. After
that (in 2017), the oral exam was skipped and replaced by individual tasks during
the semester. These individual tasks were responsible for 30% of the overall grade,
whereas the significance of the “big” group assignment was increased to 70%. In
2019, the individual tasks became related to the respective group assignments. The
reasons for these changes were that the students are much more motivated to focus
on their assigned topic and that learning here goes much deeper than the general
lecture contents. As a disadvantage, however, the familiarity with a general basis of
CI, as taught during the lectures, can be lost, since it is no longer necessary for the
grading success of the students.

4 Reasons for Adapting the Course Concepts

There are several reasons why we have changed the course concept repeatedly over
time and why, in particular, the group assignment was upgraded with respect to its
importance in student evaluation. First, this refers to the course objectives designed
to make students familiar with solving real-world problems. With regard to CI (and
many other topics), this means that it is not sufficient to teach techniques such as
modeling, simulation, and optimization to manage complex systems. Instead, it is
necessary for students to become strongly involved with related problems and apply
appropriate methods to solve them themselves without detailed supervision. Such an
approach is usually denoted as problem-based learning (PBL) and usually assumes
that students work in collaborative groups and learn by resolving complex, real-
istic problems [25]. It is supposed that this learning concept improves process skills
(teamwork, project management skills, but also autonomous learning skills and other
metacognitive skills), which we consider as particularly relevant at the Master level.
In addition, it is assumed that PBL increases the students’ motivation and engage-
ment [25]. This assumption has been confirmed by our frequent discussions with
students in the course (but also in other courses) and was also expressed in course
evaluations conducted among students (see, e.g. [26] for further details).

Another reason for a stronger focus on complex assignments can be traced back
to Bloom’s taxonomy [27]. It distinguishes the knowledge-based aspects of learning
in six levels of objectives (each represented by a characterizing noun and describing
verbs, denoted here in parentheses): 1. Knowledge (to know), 2. comprehension
(to understand, to demonstrate an understanding), 3. application (to apply, to solve
problems in new situations), 4. analysis (to analyze), 5. synthesis (to create), and 6.
evaluation (to judge). By critical examination of these levels, we found that, although
CI involves complex aspects sui generis at all levels, it is difficult to evaluate the upper

Adapting the Teaching of Computational Intelligence … 121

levels in traditional examination forms such as questions to be answered in a written
exam. Based on our experience with the various learning assessment concepts in our
CI courses, we are in line with the general experiences of universities: Mostly only
the first three to four levels of the hierarchy can be assessed in this form, whereas
questions intended for higher levels often fall back to lower levels. For instance,
the evaluation of a CI method is done by repeating arguments from the literature,
i.e. showing achievements on the level of “knowledge”. When we changed from a
written exam to an oral exam, we assumed that higher-level skills (especially levels
5 synthesis and 6 evaluation) could be better assessed in this way [26]. Although
this was partly confirmed, we were still dissatisfied with the outcomes, even taking
into account that the available time during oral exams was rather short for in-depth
assessment and that the answers result from group experiences, which do not neces-
sarily show a student’s individual achievements. For this reason we replaced the oral
exam by individual student assignments (written), which turned out to be rather easy
(see grades in Table 1 for 2017 and 2018) and did not prevent collaboration among
students who worked on the same assignments. As mentioned above, we changed
these assignments to individual assignments related to the group assignment in 2019.
Another reason for abolishing the oral exams was the provided time slot within the
two weeks for examination. With an increasing number of students, it is no longer
possible to realize oral exams with reasonable effort and without time conflicts with
other exams.

Despite such changes in relation to individual student assessments, we still believe
that the “big” group assignment is best suited to support learning objectives on all
levels of Bloom’s taxonomy. In particular, synthesis/evaluate (level 5) is strongly
supported, as the assignment requires providing an integrated solution to a complex
problem (frequently with a real-life background). Evaluation/create (level 6) is
required in various ways, e.g. with respect to the decision which solution methods to
use, how to adapt them, and in relation to comparisons with other approaches during
computational experiments.

Another reason for repeated adaptations of course concepts is given by the accred-
itation regulation of the AACSB, the Association to Advance Collegiate Schools of
,[28],whereourbusinessschoolisabouttobeaccredited.TheAACSB“pro-
vides quality assurance, business education intelligence, and professional develop-
ment services to over 1,600 member organizations and more than 800 accredited busi-
ness schools worldwide”. The assurance of learning (AoL) process specified by [29]
requires that a “school uses well-documented, systematic processes for determining
and revising degree program learning goals; designing, delivering, and improving
degree program curricula to achieve learning goals; and demonstrating that degree
program learning goals have been met”. Thus, the overall learning goals must be
broken down to individual courses and assessed on a student’s level. Since the goals
are defined in terms of students’ specific competences, the didactic concept from
teaching to grading needs to consider them. The CI course mostly focuses on the
following aspects:

122 T. Hanne and R. Dornberger

T
ab

le
1

S
ur
ve
y
st
ud
en
t
re
su
lt
s
in

C
I
co
ur
se
,
20
09
–2
01
9.

A
S
=

au
tu
m
n
se
m
es
te
r,
S
S
=

sp
ri
ng

se
m
es
te
r.
G
ra
de
s
ar
e
m
ea
su
re
d
on

a
sc
al
e
fr
om

1
to

…