Research activity
The research activities have been carried
out in various areas of the field of computational intelligence, such
as adaptive problem solving, machine learning, image processing, and
multi-agent systems. In particular, contributions have been made to
the following areas:
Biometric systems
The focus of the research was on the study
of innovative methods, algorithms and systems for the automatic
recognition of individuals in security applications. In particular,
the research focused on usable biometric systems and on touchless and
less-constrained biometric systems.
·
Usable biometric systems
In this
context, special attention was paid to the study of recent
developments in biometric recognition for Automated Border Control
systems [IJ-2, IC-3, IC-5]. In addition, the research included the
design of highly usable touch-based fingerprint recognition systems
for Automated Border Controls. To do that, computational intelligence
and image processing techniques were applied to analyze typical
problems that can affect fingerprint images. The designed system is
capable of detecting these problems, and proposing corrective actions
to the user, which can improve the quality of the obtained images
[IC-4].
·
Touchless and less-constrained biometric
systems
The research
regarded the design and development of biometric systems that
permitted a less constrained use, favoring usability. In particular,
an innovative method that can extract sweat pores from touchless
images was devised. The method used image processing techniques to
detect candidate pores and computational intelligence techniques to
discard false positives [IC-2].
Industrial informatics
The research focused on the study of image
processing methods, computational intelligence techniques, and
adaptive systems for industrial applications, with special attention
to monitoring, measuring, three-dimensional reconstruction, and
classification problems. The studied techniques have been tested on a
real-world application, the production of strand boards. Different
techniques have been developed for the quantitative and qualitative
analysis of materials and industrial processes. Special attention has
been paid to the design of specific algorithms and methods that permit
to monitor the production process of strand boards, reducing the
environmental impact [IC-1]. In addition, a prototype has been
deployed in a real factory [IC-6].
Robotics
The research regarded the creation of a
robot language that facilitates human-robot interaction [BC-1, IC-7,
IC-8, IC-9, IC-10]. This language is based on natural language and
formal logic, and employs fuzzy behaviours to translate orders into
executable code. It favors the incorporation of humans in robot teams,
helping robots to adapt to unexpected changes in the environment or in
their plans.
Electric vehicles
The research regarded the evaluation of the
possible impact of electric vehicles on Spanish power network [IJ-4,
IC-11]. To do that, a simulation technique based on queueing theory
and fuzzy logic was developed. This approach is more realistic than
previous treatments, since it models the problem taking into account
its natural imprecision, while leading to accurate descriptions of the
performance of the system. The particular problem motivating this
study was the design of a system of intelligent battery chargers
capable of adapting the charging process to guarantee the recharging
of the battery in a reasonable period of time while avoiding overload
of the power network.
Adaptive problem solving
The research regarded the study and
development of adaptive cooperative methods for the resolution of
complex problems, including NP-hard problems. The research focused on
hybrid strategies, memetic algorithms, dynamic optimization algorithms
and ant colony optimization.
·
Hybrid optimization strategies:
a framework for the design and construction of hybrid, parallel,
adaptive and cooperative metaheuristics based on the collaboration of
agents, was designed, which are able to effectively solve complex
real-world problems [IJ-8, IC-15, BC-6, NC-7, NC-8]. These strategies
employ data mining and fuzzy reasoning to incorporate previous
knowledge in controlling cooperation among agents that implement
different metaheuristics [IC-16, IC-17, BC-3, BC-4, BC-5]. To improve
its learning process, active learning techniques were applied [NC-2].
·
Memetic algorithms: memetic
algorithms are nature inspired techniques that try to overcome the
problems of genetic algorithms to find the optimum with sufficient
precision. However, memetic approaches are affected by several design
issues related to the different choices that can be made to implement
them. A multiagent-based memetic algorithm was introduced that
executes in parallel different strategies, and which can adapt its
behavior using a knowledge extraction process and fuzzy techniques.
The method has been applied to a range of problems, including plant
allocation [IJ-5], e-learning assignment [IJ-6,IC-12] or music
composition [IJ-3,IC-13,NC-4].
·
Dynamic optimization algorithms:
the resolution of problems that change during execution was studied.
To cope with the challenges that present, a cooperative strategy that
learned from previous executions was designed. In this way it can
track the optimum as it moves in the search space. In order to control
the cooperation a collection of Support Vector Machine models and a
fuzzy decision framework were used [IJ-7, NC-3, NC-6].
·
Ant colony optimization: a
complex problem that deals with the design of delivery routes in an
ecological context was studied. In this problem, a fleet of agile
environmentally friendly vehicles have to pick-up and deliver items.
However, these vehicles have a limited storage capacity, and they need
the support of larger environmentally friendly vehicles, mobile
warehouses, to increment their range. To solve it, an adapted ant
colony optimization algorithm was designed [BC-2, NC-1].
Learning with imperfect data
The research regarded the study of the
impact of imperfect data (missing, imprecise or incorrect data) on
machine learning strategies. The lack of tools that permit to
create/manage low quality datasets was detected, and a tool with such
capacities was created [NC-5, NC-9]. In addition, a technique that
permits to obtain fuzzy partitions robust to imperfect data was
designed [IC-14].