Invited Speakers
Speaker 1: Antônio de Padua Braga, Federal University of Minas Gerais, Brazil
Title: Large margin classification with graph-based models
Abstract: Autonomous learning systems have been the subject of attention of the machine learning community in recent
years. The need for methods that are capable of learning and computing in the edge has demanded models that
are less computational intensive and less dependent on user interaction and on large computational systems to
run optimization algorithms. Classification methods often require trade-off between objective functions related
to model fitness to data and to the complexity of the problem. It has been usually treated with multi-objective
or constrained optimization approaches, and trade-off accomplished according to user-defined
hyperparameters, e.g. regularization. Distance-based classification methods require less user interaction to set
parameters, since they are based on a pre-established metric and on the structure of the training data,
however, margin maximization and separation surface smoothing need also to be treated.
This presentation will describe geometrical approaches for obtaining large margin classifiers which are less user
dependent and more feasible to embedded systems implementations. The methods aim at exploring the
geometrical properties of the dataset considering only the structure of a Gabriel graph, which represents
pattern relations according to a given distance metric. Once the graph is generated, structural support vectors
(analogous to SVM's support vectors) are obtained in order to yield the final large margin solution. All
parameters of the methods are extracted from the graph or according to graph properties, so autonomous
learning can be accomplished considering that the final classifier is fully extracted from the graph. Present topics
of research include treating scalability issues associated to the learning set size and exploring further graph
properties to set implicit regularization of the separating surface.
Biography: Antonio de Padua Braga is a Professor at Department of Electronics Engineering at Federal University of Minas
Gerais, Brazil. He has been working in the field of Artificial Neural Networks since the early 1990s, when he
obtained his PhD from Imperial College London for his work on storage capacity of recurrent binary neural
networks. Since then he has co-authored many books, book-chapters, journal and conference papers and has
supervised many postgraduate students. He was a Visiting Professor at University of Alberta (Canada), and also
at Université Paris-Est (France). As a research leader he has received many grants from Brazilian researchagencies and from private companies. He has also served as Associate Editor of international journals as IEEE
Transactions on Neural Networks and Learning Systems, Neural Processing Letters and Engineering Applications
of Artificial Intelligence.
Speaker 2: Juergen Branke, University of Warwick, Coventry, United Kingdom
Title: Learning to optimise – optimal learning
Abstract: This talk discusses the relationship between machine learning and optimisation. It demonstrates that many machine learning problems are actually optimization problems, and could benefit from advances in operational research. On the other hand, the latest challenges in optimisation, such as parameter tuning, algorithm selection, Hyper heuristics or handling of uncertainty are actually closely related to machine learning. Furthermore, recent algorithmic developments such as Bayesian Optimisation very much blur the boundary between machine learning and optimisation, as they explicitly combine learning about the search space with optimisation.
Biography: Juergen Branke is Professor of Operational Research and Systems at the University of Warwick (UK). He has been working in the area of metaheuristics for over 25 years, and applied them to a wide variety of problems, including optimisation under uncertainty, dynamically changing optimisation problems, and multi-objective optimisation. Juergen has published almost 200 papers in international peer-reviewed journals and conferences. He is Editor of the ACM Transactions on Evolutionary Learning and Optimization, Area Editor of the Journal of Heuristics and the Journal on Multi-Criteria Decision Analysis, as well as Associate Editor of IEEE Transactions on Evolutionary Computation and the Evolutionary Computation Journal.
Speaker 3: Oscar Cordon, University of Granada, Spain
Title: Hybrid Intelligent Systems for Forensic Anthropology and Human Identification
Abstract: Skeleton-based forensic identification methods carried out by anthropologists, odontologists, and pathologists represent the first step in every human identification (ID) process and the victim’s last chance for identification when DNA or fingerprints cannot be applied. They include methods as biological profiling (BP), comparative radiography (CR), craniofacial superimposition (CFS), and comparison of dental records. BP involves the study of skeletal remains to find characteristic traits (age, sex, stature, and ancestry) that support determining the identity of the individual. It plays a crucial role in narrowing the range of potential matches during the process of ID, prior to the corroboration by any ID technique. CR considers the ante-mortem (AM) and post-mortem (PM) comparison of different bones and cavities (skull frontal sinuses, clavicles, patellae, …) which have been reported as useful for positive identification based on their individuality and uniqueness. CFS aims to overlay a skull with some AM images of a candidate in order to determine if they correspond to the same person.
However, practitioners still follow an observational paradigm using subjective methods introduced many decades ago; namely, oral description and written documentation of the findings obtained and the manual and visual comparison of AM and PM data. Designing systematic, automatic ad trustworthy methods to support the forensic anthropologist when applying BP, CFS and CR, avoiding the use of subjective, error-prone and time-consuming manual procedures, is mandatory to enhance forensic ID. The use of hybrid intelligent systems (in particular evolutionary algorithms, fuzzy sets and deep learning), computer vision (3D-2D image registration and image processing), and explainable machine learning is a natural way to achieve this aim. This plenary talk is devoted to present three hybrid intelligent systems for CFS, CR, and skeleton-based age-at-death assessment developed in collaboration with the University of Granada’s Physical Anthropology Lab within a fifteen years long research project. One of those systems is protected by an international patent, exploited by Panacea Cooperative Research, and is under commercialization in different countries.
Biography:Oscar Cordón was the Founder and a Leader of the Virtual Learning Center (2001-05) and the Vice President of Digital University (2015-19) with the University of Granada (UGR). He was one of the Founding Researchers with the European Centre for Soft Computing (2006-11), being contracted as Distinguished Affiliated Researcher until December 2015. He is currently a Professor with the UGR. He has been, for >25 years, an internationally recognized contributor to Research and Development Programs in fundamentals and real-world applications of computational intelligence. He has published >390 peer-reviewed scientific publications, including a research book on Genetic Fuzzy Systems (with >1400 citations in Google Scholar) and 113 JCR-SCI-indexed journal papers (67 in Q1 and 38 in D1), advised 20 Ph.D. dissertations, and coordinated 38 research projects and contracts (with an overall amount of >9M€). From June 2021, his publications had received 5439 citations (H-index=39), being included in the 1% of most-cited researchers in the world (source: Web of Science), with 14850 citations and H-index=58 in Google Scholar. He also has a granted international patent on an intelligent system for forensic identification commercialized in Mexico and South Africa.
He received the UGR Young Researcher Career Award (2004), the IEEE Computational Intelligence Society (CIS) Outstanding Early Career Award (2011, the first such award conferred), the IFSA Award for Outstanding Applications of Fuzzy Technology (2011), the National Award on Computer Science ARITMEL by the Spanish Computer Science Scientific Society (2014), the IEEE Fellow (2018), the IFSA Fellowship (2019), and the Recognition for Scientific Career and Promotion of Artificial Intelligence by the Spanish Association for Artificial Intelligence (2020). He was a member of the High-Level Expert Group that developed the Spanish R&D Strategy for Artificial Intelligence by the Spanish Ministry of Science, Innovation and Universities (2018-19). He is currently or was Associate Editor of 19 international journals. He was recognized as an Outstanding Associate Editor of IEEE Transactions on Fuzzy Systems (2008) and of IEEE Transactions on Evolutionary Computation (2019). Since 2004, he has taken many different representative positions with EUSFLAT and the IEEE Computational Intelligence Society.
His current research lines are on artificial intelligence for forensic identification (with the UGR Physical Anthropology lab and several international forensic labs and security forces) and agent-based modeling and social network analysis for marketing (with R0D Brand Consultants in projects for CAPSA, Mercedes, Jaguar-Land Rover, El Corte Inglés, Telefónica, Samsung, Coca Cola Europe, Cola Cao, WiZink, …).
.
Speaker 4: Kalyanmoy Deb, Michigan State University, USA
Title: Customized Evolutionary Optimization for Practical Problem Solving
Abstract: Evolutionary optimization methods are increasingly being used for practical problem solving tasks. It has been well established that no one optimization algorithm will be best for all problems. Even after many decades of studies, not much attention is placed in choosing or developing an appropriate optimization algorithm for a problem. In this lecture, we highlight the importance of developing a "customized" algorithm for routinely solving a problem class, rather than borrowing a standalone generalized optimization algorithm for every problem. Evolutionary optimization methods provide an ideal platform for developing a customized procedure. We shall support our argument by presenting a number of case studies involving single and multi-objective optimization problems from practice.
Biography: Kalyanmoy Deb is University Distinguished Professor and Koenig Endowed Chair Professor at Department of Electrical and Computer Engineering in Michigan State University, USA. Prof. Deb's research interests are in evolutionary optimization and their application in multi-criterion optimization, modeling, and machine learning. He was awarded IEEE Evolutionary Computation Pioneer Award for his sustained work in EMO, Infosys Prize, TWAS Prize in Engineering Sciences, CajAstur Mamdani Prize, Distinguished Alumni Award from IIT Kharagpur, Edgeworth-Pareto award, Bhatnagar Prize in Engineering Sciences, and Bessel Research award from Germany. He is fellow of IEEE and ASME. He has published over 575 research papers with Google Scholar citation of over 190,000 with h-index 127.
Speaker 5: Andries Engelbrecht, University of Stellenbosch, South Africa
Title: Multi-guide Particle Swarm Optimization for Multi- and Many-Objective Optimization
Abstract: The multi-guide particle swarm optimization (MGPSO) algorithm has originally been developed to solve multi-objective optimization problems. The MGPSO is a subswarm approach, where each subobjective is optimized by a separate swarm. In order to facilitate finding of non-dominated solutions and convergence to a Pareto-front, the particle velocity update is adapted by adding an archive guide term. The archive term serves as a mechanism to transfer knowledge about the non-dominated solutions throughout all subswarms. This talk will introduce the MGPSO algorithm and will present results to show that it performs excellently with reference to state-of-the-art approaches. The talk will then proceed to discuss the control parameters of the MGPSO, providing theoretically derived stability conditions on the control parameters to ensure that an equilibrium state is reached, to present alternative strategies to adapt the archive balance coefficient, and to analyze the importance of the control parameters. Due to the simplicity of the approach, and the use of subswarms, then MGPSO is easily scaled to many-objectives. The talk will present results to illustrate the scalability of the MGPSO in comparison with other many-objective optimization algorithms.
Biography:Prof Andries Engelbrecht received the Masters and PhD degrees in Computer Science from the University of Stellenbosch, South Africa, in 1994 and 1999 respectively. He is currently appointed as the Voigt Chair in Data Science in the Department of Industrial Engineering, with a joint appointment as Professor in the Computer Science Division, Stellenbosch University. Prior to 2019, he was appointed appointed in the Department of Computer Science, University of Pretoria (1998-2018), where he served as the head of the department (2008–2017), South African Research Chair in Artificial Intelligence (2007–2018), and Director of the Institute for Big Data and Data Science (2017–2018). His research interests include swarm intelligence, evolutionary computation, artificial neural networks, machine learning, data analytics, and the application of these artificial intelligence paradigms to data mining, data clustering, finance, and difficult optimization problems. He is author of two books, “Computational Intelligence: An Introduction” and “Fundamentals of Computational Swarm Intelligence”. He serves as an associate editor for Swarm Intelligence, IEEE Transactions on Evolutionary Computation, IEEE Transactions on Neural Networks and Learning Systems, Engineering Applications of Artificial Intelligence, and Complex and Intelligent Systems.
Speaker 6: Frédéric GUINAND, Le Havre Normandy University, France
Title: Swarms of Unmanned Aerial Vehicles
Abstract: Following big military drones and prototypes dedicated to scientific studies, lightweight drones are now very common and will probably be part of our daily life in a near future. Their agility and low cost make them very attractive for testing new usages. Their use is spreading every day more, in a growing number of domains: audiovisual, industrial inspection and maintenance, security, precision agriculture, surveillance, photogrammetry and cartography, archeology, transportation and logistics, health, leisure, disaster management, rescue, etc. Mainly, these activities rely on the use of only one UAV (fixed wing of multirotor) or on a group of remotely and individually piloted drones. However, for few years, a new challenge has been engaged by major actors of the domain: a challenge aiming at flying simultaneously many drones in groups, a challenge leading to a kind of "race for records". Thus, in 2015, Intel deployed 100 drones, establishing a word record for the largest number of UAVs flying simultaneously, a small number in comparison to the 1374 machines deployed by Ehang Egret three years later in the city of Xi'an, record broken again the same year by Intel at the occasion of the celebration of the 50th anniversary of the company. The, up-to-now, last milestone was reached in March 2021 by Genesis, the luxury vehicle division of Hyunday, with 3281 UAVs drawing in the sky the brand logo. While impressive, these demonstrations organize the flights as choregraphies. After many hours of simulation, drones are programmed and every movement of every drone is computed offline and controlled online during the demonstration. Beyond the show, for a large number of agricultural, industrial or service activities, the choreographic approach is not a possible option insofar as it does not allow the swarm to adapt to the conditions, sometimes changing, of the mission.
The expected benefit of deploying a swarm of drones is to achieve certain objectives that remain beyond the reach of a single drone, in particular when the task to be handled is constrained in time and space. These systems are in fact capable of dealing with problems arising from rapidly evolving phenomena and/or requiring a significant geographic coverage which changes under time constraints, in an original and robust manner. If the deployment of several individually piloted drones partially removes these limitations, it requires the availability of a team of pilots and drastic coordination rules to avoid problems in flight while ensuring the success of the mission. The only way of exploiting the full potential of swarms of drones is to endow them with autonomy in making decisions on the basis of local information (sensors, communications between drones). In this talk we will deal with autonomous swarms of drones, systems relying neither on a central coordination station nor on pre-programmed plans for the mission, systems in which each drone takes its own decisions based on communications with peers and/or data stemmed from embedded sensors, camera or other devices carried by the drone itself.
Biography:Dr. Frédéric Guinand received MSc degree in Computer Science in 1991, and PhD degree in Computer Science in 1995 both from Grenoble Institute of Technology (France). Engineer in a startup in 1996 for developing Internet activities, he obtained an INRIA postdoc fellowship for working at the Swiss Federal Institute of Technology of Lausanne (EPFL, Switzerland). In 1997 he joined Le Havre University as Assistant Professor and he was appointed to a Full Professor in 2005. His research interests are in the areas of dynamic graph theory, distributed and mobile computing, nature-inspired computing and collective robotics. Currently co-chair of the Complex System research axis of CNRS Normastic federation, he is also Visiting Professor at the Department of Mathematics and Natural Science at Cardinal Stefan Wyszynski University in Warsaw.
Speaker 7: Cengiz Toklu, Beykent University, Istanbul, Turkey
Title: Hybrid Algorithms. Applications to Structural Mechanics
Abstract: Metaheuristics algorithms (MAs), together with neural networks, in conscious use since several decades, formed
a real revolution in solving optimization problems in all fields of science and engineering. Versatility of MAs
made them indispensable when attacking any kind of optimization problem with all kinds of variables, with
convex and non-convex areas of definition, involving functions with undefined gradients and constraints of any
sort, with one or more objectives, with unique or non-unique solutions, etc. Their extremely high level of
popularity made that many a different type of MAs have been forwarded until now, some quite different than
others, some being difficult to differentiate from the others, based on metaphors from quite different fields like
life sciences, physics, metallurgy, sociology, etc. The number of MAs created in this way can be estimated as 200
or more. It is seen that all algorithms forwarded are successful, perhaps not for all problems, but certainly for
some.
Applications have shown that, in solving a problem, instead of using an algorithm from the start to the end, a
hybrid application, i.e., using more than one algorithm alternatively or in parallel, may be a better procedure as
far as speed and accuracy is concerned. It can be seen in the literature that this hybridization can be done not
only between MAs, but other types of algorithms as well can be considered in this procedure.
An important area of application of MAs, hybridized or not, is structural mechanics, i.e., structural design and
structural analysis. In this presentation, after giving some generalities of hybrid algorithms, their applications on
structural mechanics will be discussed based on studies some of which are carried out in our group.
Biography:Professor Toklu obtained his BS and MS degrees in Civil Engineering from Middle East Technical University,
Ankara Turkey and his doctorate from Universite de Pierre et Marie Curie (Paris VI), Paris, France. In his
professional life he directed and/or supervised numerous giant construction projects in Turkey, including a
pontoon bridge, a long span suspension bridge, a light railed transportation system, and several motorways. In
academic life he taught in several universities starting with Middle East Technical University, serving in many
cases as Department Head or Dean. Being a member of several technical and scientific international and
national organizations, he is currently affiliated to Beykent University in Istanbul, Turkey. His research interests
include application of optimization techniques to engineering, application of Artificial Intelligence concepts to
engineering, space civil engineering, nonlinear analysis of structures, engineering education and constructionscheduling. He is the author of several books, book chapters and scientific articles. He has organized many
congresses and served as keynote speaker in many international meetings. Dr. Toklu is the developer of the
method “Total Potential Optimization using Meta-heuristic Algorithms (TPO/MA)” that gave way to the method
Finite Element Method with Energy Minimization (FEMEM) which is shown to be more successful than classical
methods in analyzing non-linear structural systems, under-constrained structures, unstable structures,
degenerate structures and structures with non-unique deformed shapes. His current research is on producing
lunar soil simulant and lunar construction materials including lunar bricks, lunar concrete, and the like.
Speaker 8: Günther Raidl, Technische Universität Wien, Austria
Title: Combinatorial Optimization Meets (Reinforcement) Learning
Abstract: The machine learning boom of the last years also led to interesting new developments in the area of combinatorial optimization. Classical optimization techniques for approaching hard combinatorial problems include many based on tree search, frequently in combination with linear programming or constraint propagation, but also various kinds of metaheuristics. While end-to-end machine learning approaches are still far from replacing these classical techniques, it has been recognized that the latter may benefit from incorporating learning components for certain purposes. One may say the aim is to ``learn how to better optimize''.
This talk will give an overview on some promising recent developments in this direction. For example for branch-and-bound, approaches have been proposed that learn better performing branching and node selection strategies. In beam search, guidance heuristics may be learned that yield better results than leading manually crafted heuristics. In the area of metaheuristics, we will look at large neighborhood search approaches where the construction of the neighborhoods is learned. Some of these methods rely on imitation or supervised learning approaches where labeled training data or some powerful other method to learn from need to be available. More versatile may be methods based on reinforcement learning principles, on which we will also have a look at.
Biography:Günther Raidl is Professor at the Institute of Logic and Computation, TU Wien, Austria, and member of the Algorithms and Complexity Group. He received his PhD in 1994 and completed his habilitation in Practical Computer Science in 2003 at TU Wien. In 2005 he received a professorship position for combinatorial optimization at TU Wien.
His research interests include algorithms and data structures in general and combinatorial optimization in particular, with a specific focus on metaheuristics, mathematical programming, intelligent search methods, and hybrid optimization approaches. His research work typically combines theory and practice for application areas such as scheduling, network design, transport optimization, logistics, and cutting and packing.
Günther Raidl is associate editor for the INFORMS Journal on Computing and the International Journal of Metaheuristics and at the editorial board of several journals including Algorithms, Metaheuristics and the Evolutionary Computation. He is co-founder and steering committee member of the annual European Conference on Evolutionary Computation in Combinatorial Optimization (EvoCOP). Since 2016 he is also faculty member of the Vienna Graduate School on Computational Optimization.
Günther Raidl has co-authored a text book on hybrid metaheuristics and over 180 reviewed articles in scientific journals, books, and conference proceedings. Moreover, he has co-edited 13 books and co-authored one book on hybrid metaheuristics. More information can be found at http://www.ac.tuwien.ac.at/raidl.
Speaker 9: Yukio Ohsawa, The University of Tokyo, Japan
Title: Elicitation of Feature Concepts as Data Federative Innovation Literacy
Abstract: Since 2000, the speaker has initiated and embodied Chance
Discovery, a subdomain of data science, meaning to detect and explain
a chance, that is a piece of high-utility information as part of data
about events meaningful for human's decision making. At that time, he
thought a network of networks can be the essential model for
representing the latent dynamics where an edge between networks is
linked to a chance. Then, he extended the methods of chance discovery
for explaining the utility of datasets, via the analogy between an
event as the base and the metadata of a dataset as the target. As well
as the base problem of chance discovery, that is to explain the
utility of information about an event considering its relation to
other events, the utility of a dataset as the target goal could be
explained its relation to other datasets. However, he found
information obtained from a dataset created by a combination of
different but connectable (sharing attributes and/or purpose of using)
datasets is essentially hard to interpret because the same analysis
models as of the original datasets can not be applied directly due to
the difference in the requirements of data user(s). Thus, it comes to
be an important problem to elicit a new "feature concept" for target
data. A feature concept is a model of the concept to be retrieved from
data that can not be represented by a simple feature such as a single
variable but can be by a conceptual illustration. Decision trees,
Clusters, and even deep neural networks can be positioned as examples
of feature concepts. A useful feature concept for satisfying a
requirement of a data user has been elicited via creative
communication using Data Jackets among data providers, data users, and
other stakeholders in the market of data. In this keynote, the history
of chance discovery and data-jacket-based design of creative
communication is reviewed with some cases of application -- marketing,
detection of earthquake precursors, suppression of COVID-19 spreading
risk, etc, cases and highlight the feature factors elicited and used
in these examples. An essential message here is that sharing and
using/reusing feature concepts is literacy for data-federative
innovations.
Speaker 10: Marco Dorigo, Université Libre de Bruxelles, Belgium
Title: Improving controllability, robustness and security of robot swarms
Abstract: In the first part of the talk, I will discuss the concept of a mergeable nervous system for robot swarms and present experimental results that show how it can be used to design and implement swarms of robots that are easier to control and that can self-repair. In the second part, I will present some recent results of research aimed to improve security in robot swarms via the use of Merkle trees and of the blockchain.
Biography: Marco Dorigo received the Laurea, Master of Technology, degree in industrial technologies engineering in 1986, and the Ph.D. degree in electronic engineering in 1992 from the Politecnico di Milano, Milan, Italy, and the title of Agr{\'e}g{\'e} de l’Enseignement Sup{\'e}rieur, from ULB, in 1995. From 1992 to 1993, he was a Research Fellow at the International Computer Science Institute, Berkeley, CA. In 1993, he was a NATO-CNR Fellow, and from 1994 to 1996, a Marie Curie Fellow. Since 1996, he has been a tenured Researcher of the FNRS, the Belgian National Funds for Scientific Research, and co-director of IRIDIA, the artificial intelligence laboratory of the ULB. He is the inventor of the ant colony optimization metaheuristic. His current research interests include swarm intelligence, swarm robotics, and metaheuristics for discrete optimization. He is the Editor-in-Chief of Swarm Intelligence, and an Associate Editor or member of the Editorial Boards of many journals on computational intelligence and adaptive systems. Dr. Dorigo is a Fellow of the AAAI, EurAI, and IEEE. He was awarded the Italian Prize for Artificial Intelligence in 1996, the Marie Curie Excellence Award in 2003, the Dr. A. De Leeuw-Damry-Bourlart award in applied sciences in 2005, the Cajastur International Prize for Soft Computing in 2007, an ERC Advanced Grant in 2010, the IEEE Frank Rosenblatt Award in 2015, and the IEEE Evolutionary Computation Pioneer Award, awarded in 2016.