The main aim of this tutorial is to educate a new generation of type-2 fuzzy systems researchers. The main audience of this tutorial will be graduate students, young researchers, engineers or researchers who would like to start working in the area of type-2 fuzzy systems. The tutorial will be a two part tutorial, where the first part of the tutorial (2 hours) will concentrate on the theoretical basis and definitions of type-2 fuzzy sets and systems while giving a complete coverage of interval type-2 fuzzy systems. The tutorial will use different means of simplifying and explaining the subject through different examples. The second part of the tutorial (3 hours) will be split into two parts addressing the practical aspects and the future directions of type-2 fuzzy systems. The practical aspects part of the tutorial will present the necessary algorithms needed to develop complete type-2 fuzzy systems applications. This part of the tutorial will present several applications of type-2 fuzzy systems. We will also explain when to use type-2 fuzzy systems and will guide the audience through the steps needed to design a type-2 fuzzy system. Several interactive examples will be presented to the audience to make sure that the audience can start to confidently design type-2 fuzzy systems. The future directions part will explore future directions of type-2 fuzzy systems explaining what the ‘hot topics’ are thus allowing, for example, PhD students to see where they might take their work. It will include a short overview of the progress with generalised type-2 fuzzy systems

The tutorial presents the novel theory of fuzzy networks, its application for some case studies and the Matlab implementation of some of its methods. The first tutorial section discusses complexity as a systemic feature and the ability of fuzzy systems to handle different attributes of complexity. Section 2 reviews several types of fuzzy systems in the context of systemic complexity, including systems with single, multiple and networked rule bases. Section 3 introduces formal models for fuzzy networks such as Boolean matrices, binary relations, block schemes and topological expressions. Section 4 presents basic operations on nodes in fuzzy networks, including merging and splitting in horizontal, vertical and output context. Section 5 discusses structural properties of basic operations such as associativity of merging and variability of splitting in horizontal, vertical and output context. Section 6 describes advanced operations on nodes in fuzzy networks, including node transformation for input augmentation, output permutation and feedback equivalence, as well as node identification in horizontal, vertical and output merging. Section 7 shows the application of the theoretical results from Sections 3-6 in feedforward fuzzy networks with single or multiple levels and layers. Section 8 illustrates the application of the theoretical results from Sections 3-6 in feedback fuzzy networks with single or multiple local and global feedback. Section 9 evaluates fuzzy networks by means of assessment of structural complexity, composition of hierarchical fuzzy systems, decomposition of standard fuzzy systems, indicators of model performance and applications for case studies. Section 10 highlights the theoretical significance, the application areas and the methodological impact of fuzzy networks. The last tutorial section presents the Matlab programs from the Fuzzy Networks Toolbox developed by Nedyalko Petrov – a PhD student from the University of Portsmouth, UK.

Dynamic systems assume different functioning modes in the course of time. In statistical Pattern Recognition (PR), historical patterns about system functioning modes are divided into groups of patterns, called classes. The classification of patterns can be achieved using membership functions which determine the membership value of a new pattern to a class, i.e. functioning mode.

Patterns are static or dynamic. A static pattern is represented by a point in the feature space. A dynamic pattern is represented by a multidimensional trajectory in the feature space. In this case, the feature space has an added dimension which is the time. Classes can be also static or dynamic. Static classes are based on stationary patterns. These classes are represented by restricted areas, formed by similar patterns, in the feature space. Thus, the way in which patterns arrive is irrelevant to their membership values. Therefore, the classifier's parameters remain unchanged with the time. However, most of data issued from the real world are non stationary. In this case, classes become dynamic and their characteristics change in the course of time. Thus, the classes’ membership functions must be adapted to take into account these temporal changes. This requires an adaptive classifier with a mechanism for adjusting its parameters over the time.

In this tutorial, the principles of dynamic PR methods will be presented and compared according to some meaningful criteria. Then, their application to solve some of real world problems will be discussed to show their interest according to the static classifiers.

The goal of this tutorial is to present advanced soft computing methods and tools for decision making in the areas of evaluation, optimization, and comparison of complex systems and alternatives. The tutorial includes three parts: theory, methods, and applications.

The first part of tutorial presents the fundamental concepts of continuous preference logic: andness, orness, hard and soft partial conjunction, hard and soft partial disjunction, partial absorption, and the interaction of formal logic concepts and semantic concepts that are fundamental in human reasoning. We use graded logic functions to develop aggregators that provide adjustable degrees of conjunction, disjunction, absorption, and relative importance. The main goal of these mathematical models is to be fully consistent with observable properties of human reasoning.

The second part of tutorial presents the Logic Scoring of Preference (LSP) method for building industrial strength decision models for evaluation, comparison, optimization and selection of complex systems. The LSP method includes techniques for training of aggregators, systematic development of evaluation criteria, sensitivity analysis, reliability analysis, optimization, cost-preference analysis, and verbalization of evaluation results. We also present software tools that support these techniques.

The third part of tutorial surveys areas of application of the LSP methodology, and presents three convincing case studies: (1) decision making in space management based on LSP suitability maps, (2) medical decision making based on LSP models of disease severity and patient disability, and (3) the use of LSP method to compare and select complex software systems.

Information not available.

More and more information about processes is recorded in the form of event logs. Enterprise information systems, medical devices, RFID-based systems, web services, etc. are all collecting events related to the processes they support. This data explosion allows for the analysis of a wide variety of operational processes. Process mining provides a versatile and extendible way to analyze such processes. Using process mining techniques it is possible to extract different types of models from event logs, e.g., the construction of process models, organizational models, performance models, etc. State-of-the-art process mining techniques are able to discover complex processes thus enabling organizations to understand and improve the way in which people work.

Process mining is not limited to discovery. Using conformance checking techniques existing models can also be compared with reality and enhanced with additional information, e.g., indicating bottlenecks in a process.

Many vendors claim to offer support for Business Intelligence (BI). Unfortunately, these BI tools are not intelligent at all. Moreover, these tools require input data of a particular type and a predefined model. Process mining overcomes these limitations and makes it possible to extract new knowledge from information systems in a truly intelligent way. Hence, process mining techniques are being adopted by commercial tools such as BPM|one, Futura Reflect, ARIS PPM, Fujitsu Interstage, etc. Moreover, the IEEE recently established a task force on process mining.

This tutorial aims to provide an overview of process mining techniques and, using many real-life examples, it will be shown how particular techniques can be applied and what kind of insights such analyses provide.

Prof.dr.ir. Wil van der Aalst is a full professor of Information Systems at the Technische Universiteit Eindhoven (TU/e) having a position in both the Department of Mathematics and Computer Science and the Department of Technology Management. Currently he is also an adjunct professor at Queensland University of Technology (QUT) working within the BPM group there. His research interests include workflow management, process mining, Petri nets, business process management, process modeling, and process analysis. Wil van der Aalst has published more than 115 journal papers, 15 books (as author or editor), 230 refereed conference/workshop publications, and 40 book chapters. Many of his papers are highly cited (he has an H-index of more than 69 according to Google Scholar, making him the Dutch computer scientist with the highest H-index) and his ideas have influenced researchers, software developers, and standardization committees working on process support. He has been a co-chair of many conferences including the Business Process Management conference, the International Conference on Cooperative Information Systems, the International conference on the Application and Theory of Petri Nets, and the IEEE International Conference on Services Computing. He is also editor/member of the editorial board of several journals, including the Distributed and Parallel Databases, the International Journal of Business Process Integration and Management, the International Journal on Enterprise Modelling and Information Systems Architectures, Computers in Industry, Business & Information Systems Engineering, IEEE Transactions on Services Computing, Lecture Notes in Business Information Processing, and Transactions on Petri Nets and Other Models of Concurrency. He is also a member of the Royal Holland Society of Sciences and Humanities (Koninklijke Hollandsche Maatschappij der Wetenschappen).

For more information about his work visit: www.workflowpatterns.com, www.workflowcourse.com, www.processmining.org, www.yawl-system.com, www.wvdaalst.com.

Image registration (IR) is a fundamental task in computer vision used to finding either a spatial transformation or a correspondence among two or more images acquired under different conditions. During the last few decades, IR has been established as a very active research issue with applications to many real-world problems ranging from remote sensing to medical imaging, artificial vision, and computer-aided design. Several aspects cause difficulties for the success of the optimization process of traditional IR methods, which are prone to be trapped in local minima. Nevertheless, evolutionary algorithms have successfully tackled IR problems thanks to their global optimization nature and their capability to perform robust search in complex and ill-defined problems.

The aim of the current tutorial is to review the first and last developments on the evolutionary IR field. We will especially emphasize its real-world applicability by reporting a broad experimental study comparing the performance of fourteen methods facing two of the most relevant application domains: medical IR (considering real magnetic resonance images and computerized tomographies) and range IR for 3D modeling. Finally, we will show the most recent outcomes of a coordinated project between the European Centre for Soft Computing and the Physical Anthropology Lab of the University of Granada (Spain) where evolutionary IR is being applied to forensic identification. The results obtained in several real-world identification cases solved in collaboration with the Spanish Scientific Police will be reported.

Dr. Oscar Cordón is Principal Researcher at the European Centre for Soft Computing (Mieres, Spain). He has been an assistant professor since 1995 and associate professor since 2001 at the University of Granada (Spain). He was the founder and leader of that University?s Virtual Learning Center (CEVUG) between 2001 and 2005, and was awarded with the Young Researcher Career Award in 2004. In December 2008, he received the Full Professor accreditation from the Spanish Quality Evaluation Agency. He has been for 15 years an internationally recognized contributor to R&D Programs in fuzzy systems, evolutionary algorithms, and ant colony optimization. He has published 230 peer-reviewed scientific publications (including a research book on genetic fuzzy systems and 48 JCR-SCI-indexed journal papers), advised 11 PhD dissertations, participated in 28 research projects and contracts (being the coordinator in 18 of them), and co-edited 7 special issues in international journals and 3 research books. By February, 20, 2010, his publications had received 1186 citations, carrying an h index of 19, and being included in the 1% of most cited researchers in the world. He is in the Editorial Board of 8 international journals (being recognized as IEEE TFS Outstanding AE in 2008) and is reviewer for more than 30 international journals. Since 2004, he has taken many different representative positions in Eusflat and IEEE CIS, currently being a member of the IEEE CIS AdCom (term 2010-2010).

Dr. Sergio Damas is Associate Researcher at the European Centre for Soft Computing (Mieres, Spain). He has been an assistant professor at the University of Granada (Spain) since 1995 and permanent professor since 2007. In January 2008, he received the Associate Professor accreditation from the Spanish Quality Evaluation Agency. He has published more than 50 peer-reviewed scientific publications, including 8 SCI-JCR-indexed journal papers in prestigious journals such as ACM Computing Surveys, Information Sciences, Pattern Recognition Letters, Image and Vision Computing, and Soft Computing. He has edited a special issue in the International Journal of Approximate Reasoning (IJAR) on ?Soft Computing in Image Processing?, also indexed in the SCI-JCR. He has organized different international workshops and special sessions in national and international conferences on soft computing and computer vision. He has supervised a PhD dissertation on the latter topic. He has participated in 10 national and European research projects and contracts.

The tutorial gives a systematic view, based on the experience from The Dow Chemical Company, of the key issues for applying and creating value with Computational Intelligence (CI). The competitive advantages and the benefits of using CI in industry are defined as well as the criteria for successful practical applications. Special attention is given to marketing the approach to technical and non technical audience. The implementation issues of several computational intelligence techniques, such as neural networks, support vector machines, genetic programming, and particle swarm optimizers are discussed.

An integrated methodology, which uses the discussed methods for variable selection, data compression, and robust empirical model building is proposed and illustrated with successful industrial applications in the area of process monitoring, optimization, data mining, and new product design. The open technical and non-technical issues of applied CI are addressed and selected areas of future research are recommended.

The academic attendees will benefit from the tutorial by better awareness of the economic impact of CI, understanding the industrial needs, and learning about the details of successful industrial applications. The industrial attendees will benefit from the tutorial by understanding the different sources of value creation of CI, the proposed application strategy for introducing and leveraging the technology in an industrial setting, and the shared experience from real-world applications.

Arthur Kordon is a Data Mi nin g & Modeling Leader in the Data Mi nin g & Modeling Group, The Dow Chemical Company in Freeport , Texas , USA . He is an internationally recognized expert in applying computational intelligence technologies in industry. Dr. Kordon has successfully introduced several novel technologies for improved manufacturing and new product design, such as robust inferential sensors, automated operating discipline, accelerated fundamental model building, etc. His research interests include application issues of computational intelligence, robust empirical modeling, intelligent process monitoring and control, and data mi nin g. He has published more than 60 papers, one book and nin e book chapters in the area of applied computational intelligence and advanced control.  Dr. Kordon is a member of the Technical Committee on Evolutionary Computation of IEEE Computational Intelligence Society.

Dr. Kordon holds a Master of Science degree in Electrical Engineering from the Technical University of Varna, Bulgaria in 1974 and a Ph.D. degree in Electrical Engineering from the Technical University of Sofia, Bulgaria in 1990.

Parallel and distributed computing can be used in the design and implementation of metaheuristics (e.g. evolutionary algorithms) to speedup the search, improve the quality of the obtained solutions, improve the robustness of the obtained solutions, and solve large scale problems.

From the algorithmic design point of view, we will present the main parallel models for metaheuristics (algorithmic level, iteration level, solution level). We will address also: parallel hybrid models with exact methods, parallel models for multi-objective optimization, and illustrations solving large challenging applications in telecommunications, logistics/transportation and bioinformatics.

From the implementation point of view, we here concentrate on the parallelization of metaheuristics on general-purpose parallel and distributed architectures, since this is the most widespread computational platform. The rapid evolution of technology in terms of processors (GPUs, multi-core), networks, and architectures (GRIDs, clusters) make those architectures very popular nowadays.

Finally, some software frameworks for parallel metaheuristics such as PARADISEO are presented. Those frameworks allow the design of parallel and hybrid metaheuristics for mono-objective and multi-objective optimization, and the transparent implementation on different parallel and distributed architectures using adapted middleware.

Prof. El-ghazali Talbi received the Master and Ph.D degrees in Computer Science, both from the Institut National Polytechnique de Grenoble in France. Then he became an Associate Professor in Computer Sciences at the University of Lille (France). Since 2001, he is a full Professor at the University of Lille and the head of the optimization group of the Computer Science laboratory (LIFL). His current research interests are in the field of multi-objective optimization, parallel algorithms, metaheuristics, combinatorial optimization, cluster and grid computing, hybrid and cooperative optimization, and application to logistics/transportation, bioinformatics and networking.

Professor Talbi has to his credit more than 100 publications in journals, chapters in books, and conferences. He is the co-editor of three books. He was a guest editor of more than 10 special issues in different journals (Journal of Heuristics, Journal of Parallel and Distributed Computing, European Journal of Operational Research, Theoretical Computer Science, Journal of Global Optimization). He is the head of the INRIA Dolphin project and the bioinformatics platform of the Genopole of Lille. He has many collaborative national, European and international projects.

He is the co-founder and the coordinator of the research group dedicated to Metaheuristics: Theory and Applications (META). He is the founding co-chair of the NIDISC workshop on nature inspired computing (IEEE/ACM IPDPS). He served in different capacities on the programs of more than 100 national and international conferences. He is also the organizer of many conferences (e.g. EA'2005, ROADEF'2006, META'2008, IEEE AICCSA'2010).

The field of Evolutionary Computation has experienced tremendous growth over the past 20 years, resulting in a wide variety of evolutionary algorithms and applications. The result poses an interesting dilemma for many practitioners in the sense that, with such a wide variety of algorithms and approaches, it is often hard to se the relationships between them, assess strengths and weaknesses, and make good choices for new application areas.

This tutorial is intended to give an overview of a general EC framework that can help compare and contrast approaches, encourages crossbreeding, and facilitates intelligent design choices. The use of this framework is then illustrated by showing how traditional EAs can be compared and contrasted with it, and how new EAs can be effectively designed using it.

Finally, the framework is used to identify some important open issues that need further research.

Kenneth A. De Jong received his Ph.D. in computer science from the University of Michigan in 1975. He joined George Mason University in 1984 and is currently a Professor of Computer Science, head of the Evolutionary Computation laboratory, and the Associate Director of the Krasnow Institute. His research interests include evolutionary computation, machine learning, and adaptive systems. He is currently involved in research projects involving the development of new evolutionary algorithm (EA) theory, the use of EAs as heuristics for NP-hard problems, and the application of EAs to the problem of learning task programs in domains such as robotics, diagnostics, navigation and game playing. He is also interested in experience-based learning in which systems must improve their performance while actually performing the desired tasks in environments not directly their control or the control of a benevolent teacher. He is an active member of the Evolutionary Computation research community and has been involved in organizing many of the workshops and conferences in this area. He is the founding editor-in-chief of the journal Evolutionary Computation (MIT Press), and a member of the board of the ACM SIGEVO. He is the recipient of an IEEE Pioneer award in the field of Evolutionary Computation and a lifetime achievement award from the Evolutionary Programming Society.

This tutorial will discuss the state-of-the-art of hyper-heuristics and related approaches that share the common goal of automating the design and adaptation of heuristic methods to solve hard computational search problems.  It is concerned with search methods which explore a search space of heuristics (rather than a search space of potential solutions to a problem). The goal is that hyper-heuristics will lead to more general systems that are able to automatically operate over a wider range of problem domains than is possible today, without manually having to customise the search, or its parameters, for each particular problem domain. We have identified two main types of approaches to this challenge: automated heuristic selection, and automated heuristic generation. In heuristic selection the idea is to automatically come up with a combination of fixed pre-existing simple heuristics or neighbourhood structures to solve the problem at hand; whereas in heuristic generation the idea is to create new heuristics (or heuristic components) suited to a given problem or class of problems. This latter approach is typically achieved by combining, through the use of genetic programming for example, components or building-blocks of human designed heuristics. Although the term hyper-heuristic is relatively new, the ideas have actually been around for over 40 years. This tutorial will go over the intellectual roots and origins of both automated heuristic selection and generation, before discussing work carried out to date in these two directions and then focusing on some observations and promising research directions.

Gabriela Ochoa is a Senior Research Fellow with the Automated Scheduling, Optimisation and Planning (ASAP) research group in the School of Computer Science at the University of Nottingham, UK, since October 2006. She received a BSc degree in computer engineering in 1991 and a MRes degree in computer science in 1996 from the Department of Computer Science at the University Simon Bolivar in Caracas, Venezuela. In 2001, she completed a PhD degree in computer science and artificial intelligence from the University of Sussex, UK, working with the notion of error threshold in evolutionary algorithms. Her current research interest lies on search methodologies and machine learning, with emphasis on evolutionary algorithms, meta-heuristics and hyper-heuristics. She has also been involved in foundational studies of evolutionary algorithms and fitness landscape analysis. She has published several papers in well known international conferences and journals, has organized several workshops and special sessions, has been a member of the program committees in major international conferences, and has refereed for recognized journals in these fields.

Matthew Hyde received the PhD in Computer Science from the University of Nottingham, U.K., in 2009. He is currently a Research Fellow within the Automated Scheduling, Optimisation, and Planning (ASAP) Research Group at Nottingham. His research interests include evolutionary computation, hyper-heuristics, metaheuristics, and operational research. He has worked on two EPSRC (Engineering and Physical Sciences Research Council) funded projects. The first was to investigate genetic programming as a hyper-heuristic, and he is currently working within a 2.6M project which aims to investigate methodologies suitable to automate the heuristic design process. He has served on the program committee for the 2007 and 2010 Conference on Evolutionary Computation, and has reviewed papers for four international journals. He has published five refereed papers and two book chapters on the subject of automatic heuristic generation.

Professor Edmund Burke is Dean of the Faculty of Science at the University of Nottingham and he leads the Automated Scheduling, Optimisation and Planning (ASAP) Research Group in the School of Computer Science. He is a member of the EPSRC Strategic Advisory Team for Mathematics. He is a Fellow of the Operational Research Society and the British Computer Society and he is a member of the UK Computing Research Committee (UKCRC). Prof. Burke is Editor-in-chief of the Journal of Scheduling, Area Editor (for Combinatorial Optimisation) of the Journal of Heuristics, Associate Editor of the INFORMS Journal on Computing, Associate Editor of the IEEE Transactions on Evolutionary Computation and a member of the Editorial Board of Memetic Computing. Prof. Burke has played a leading role in the organisation of several major international conferences in his research field in the last few years. He has edited/authored 14 books and has published over 200 refereed papers. He has been awarded 54 externally funded grants worth over 13M from a variety of sources including EPSRC, ESRC, BBSRC, EU, Research Council of Norway, East Midlands Development Agency, HEFCE, Teaching Company Directorate, Joint Information Systems Committee of the HEFCs and commercial organisations. This funding portfolio includes being the Principal Investigator on an EPSRC Science and Innovation award of 2M, an EPSRC grant of 2.6M to investigate the automation of the heuristic design process and an EPSRC platform renewal grant worth over 1M.

One of the most commonly-used metaphors to describe the process of heuristic search methods in solving combinatorial optimisation problems is that of a fitness landscape. The landscape metaphor appears commonly in work related to evolutionary algorithms; the search space can be regarded as a spatial structure where each point (solution) has a height (objective function value) forming a landscape surface. In this scenario, the search process would be an adaptive-walk over a landscape that can range from having many peaks of high fitness flanked by cliffs falling to profound valleys of low fitness, to being smooth, with low hills and gentle valleys.

Combinatorial landscapes can be seen as a graph whose vertices are the possible solutions. If two solutions can be transformed into each other by a suitable operator move, then we can trace an edge between them. The resulting graph, with an indication of the fitness at each vertex, is the fitness landscape. The study of the fitness landscape consists in analysing this graph.

This tutorial will give an overview of the origins of the fitness landscape metaphor, and will cover the alternative ways to define fitness landscapes in evolutionary computation. The two main geometries: multimodal and neutral landscapes will be considered. Furthermore, the relationship between problem hardness and fitness landscape metrics, and the Local Optima Network properties, studied in recent work, will be deeply analysed. Finally, the tutorial will conclude with a brief survey of open questions and recent research directions on fitness landscapes.

Sébastien Verel is an associate professor in Computer Science at the University of Nice Sophia-Antipolis, France, since 2006. He received a BSc in mathematics from the University of Caen, France, in 1999, a MSc degree and a PhD in computer science from the University of Nice Sophia-Antipolis, France, in 2002 and 2005, respectively. His PhD work was related to fitness landscape analysis in combinatorial optimization. He is a member of the Institut des Systèmes Complexes Paris Ile-de-France, and is currently an invited researcher in DOLPHIN Team at INRIA Lille Nord Europe, France. His research interests are in the theory of evolutionary computation, cognitive science modelling, and complex systems. He has published several papers in well known international conferences and journals, has been a member of the program committees in major international conferences, and has refereed for recognized journals in these fields.

Gabriela Ochoa is a Senior Research Fellow with the Automated Scheduling, Optimisation and Planning (ASAP) research group in the School of Computer Science at the University of Nottingham, UK, since October 2006. She received a BSc degree in computer engineering in 1991 and a MRes degree in computer science in 1996 from the Department of Computer Science at the University Simon Bolivar in Caracas, Venezuela. In 2001, she completed a PhD degree in computer science and artificial intelligence from the University of Sussex, UK, working with the notion of error threshold in evolutionary algorithms. Her current research interest lies on search methodologies and machine learning, with emphasis on evolutionary algorithms, meta-heuristics and hyper-heuristics. She has also been involved in foundational studies of evolutionary algorithms and fitness landscape analysis. She has published several papers in well known international conferences and journals, has organized several workshops and special sessions, has been a member of the program committees in major international conferences, and has refereed for recognized journals in these fields.

The biological immune system (BIS) has multi-layered architecture with defenses at all levels to protect the body from invading pathogens such as viruses and bacteria. It uses feature extraction, memory, diversity and associative retrieval to solve recognition and classification tasks. These remarkable information-processing abilities of the immune system has inspired an emerging field, sometimes referred to as the Immunological Computation, Immuno-computing or Artificial Immune Systems (AIS) that extracts ideas from BIS to develop computational tools for solving science and engineering problems.

Over the last two decades, there has been an increased interest in immuno-inspired techniques and their applications. In general, some of such models are intended to describe immunological processes for a better understanding of the dynamical behavior of the BIS in the presence of antigens. On the other hand, immunity-based models have been developed in an attempt to solve wide variety of real-world problems. In particular, there exist a number of applications in pattern recognition, fault detection, computer security; also other applications currently being explored in science and engineering problem domain. This tutorial will cover the latest advances in Immunological approaches and a few real-world applications along with software demo.

Dr. Dasgupta's research interests broadly span the areas of scientific computing, tracking real-world problems through interdisciplinary cooperation. His areas of special interests include Artificial Immune Systems, Genetic Algorithms, Neural Networks, multi-agent systems and their applications. He published more than 185 research papers in book chapters, journals, and international conferences. Dr. Dasgupta published the first book in the field on Artificial Immune Systems and Their Applications, in 1998 (available in Russian). In 2008, he co-authored a Text book on Immunological Computation: Theory and Applications, CRC press. Dr. Dasgupta also published two edited volumes and co-edited several conference proceedings over the last 15 years. He is the founding chair of IEEE Task force on Artificial Immune Systems. In 2007, he received the University of Memphis “Dunavant Professorship” for 3 years as a special recognition of his research. Other awards include 3 Best Paper Awards at International Conferences; Sigma Xi Research Paper Award, Distinguished Research Award (CASDRA) twice, Early Career Research Award (ECRA) from the University of Memphis. Dr. Dasgupta regularly serves as panelist, keynote speaker and offer tutorials in leading conferences, and has given more than 50 invited talks in different universities and industries. His research lab regularly updates AIS Bibliography and publishes on the web (available at: http://ais.cs.memphis.edu/).

Evolutionary game theory has been introduced essentially by biologists in the seventies and has immediately diffused into economical and sociological circles. Today, it is a main pillar of the whole edifice of game theory and widely used both in theory and in applications. This tutorial aims at presenting evolutionary game theory in an easy, yet rigorous way and to relate it with other approaches in game theory. The material presented does not require a previous acquaintance with standard game theory: these fundamentals will be developed in the first part of the tutorial, which is self-contained. In the second part the main concepts of the evolutionary and dynamical approach will be introduced, namely the concept of an evolutionarily stable strategy and the replicator dynamics. The analogies between Nash equilibria, evolutionarily stable strategies, and rest points of the dynamics will be explained. All the concepts will be illustrated using simple well known paradigmatic games such as the Prisoner's Dilemma, Hawks anbd Doves and coordination games among others. Finally, some recente trends in evolutionary, network-based game theory will be hinted at.

Marco Tomassini is a professor of Computer Science at the Information Systems Department of the University of Lausanne, Switzerland. He graduated in physical and chemical sciences and got a Doctor's degree in theoretical chemistry from the University of Perugia, Italy, working on computer simulations of condensed matter systems. His current research interests are centered around the application of biological ideas to artificial systems. He is active in evolutionary computation, especially spatially structured systems, genetic programming, and the structure of program search spaces. He is also interested in machine learning, evolutionary games, and the dynamical properties of networked complex systems. He has been Program Chairman of several international events and has published many scientific papers and several authored and edited books in these fields.

We are living in the peta-byte era. We have larger and larger data to analyze, process and transform into useful answers for the domain experts. Robust data mining tools, able to cope with petascale volumes and/or high dimensionality producing human-understandable solutions are key on several domain areas. Genetics-based machine learning (GBML) techniques are perfect candidates for this task. Recent advances in representations, learning paradigms, and theoretical modeling have made them very suitable for large scale data analysis. If evolutionary learning techniques aspire to be a relevant player in this context, they need to have the capacity of processing these vast amounts of data within reasonable time. Moreover, massive computation cycles are getting cheaper and cheaper every day, giving access to unprecedented computational resources on the edge of petascale computing. Several topics are interlaced in these two requirements such as (1) having the proper learning paradigms and knowledge representations, (2) understanding them and knowing when are they suitable for the problem at hand, (3) using efficiency enhancement techniques, and (4) transforming and visualizing the produced solutions to give back as much insight as possible to the domain experts.

This tutorial will try to shed light to the above mentioned questions, following a roadmap that starts exploring what large scale means, and why large is a challenge and opportunity for GBML methods. As we will show later, opportunity has multiple facets: Efficiency enhancement techniques, representations able to cope with large dimensionality spaces, scalability of learning paradigms, and alternative programming models, each of them helping to make GBML very attractive for large-scale data mining. Given these building blocks, we will continue to unfold how can we model the scalability of GBML systems targeting a better engineering effort that will make embracing large datasets routine. Finally, we will illustrate how all these ideas fit by reviewing real applications of GBML systems and what further directions will require serious consideration.

Dr. Jaume Bacardit received his Ph.D. in 2004 from the Ramon Llull University in Barcelona, Spain. His thesis studied the adaptation of the Pittsburgh approach of Learning Classifier Systems (LCS) to Data Mining tasks. In 2005 he joined the University of Nottingham, UK as a postdoctoral researcher to work on the application of LCS to data mine large-scale bioinformatics datasets and extract interpretable explanations from the learning process. In 2008 he was appointed as a Lecturer in Bioinformatics at the University of Nottingham. This is a joint post between the schools of Biosciences and Computer Science with the aim of developing interdisciplinary research at the interface of both disciplines. In the School of Computer Science he is part of the ASAP research group. In the School of Biosciences he is part of the Multidisciplinary Centre for Integrative Biology (MyCIB). His research interests include the application of Learning Classifier Systems and other kinds of Evolutionary Learning to data mine large-scale challenging datasets and, in a general sense, the use of data mining and knowledge discovery for biological domains. He is co-chair of the International Workshop on Learning Classifier Systems since 2007 and in 2009 he was the chair of the Genetics-Based Machine Learning Track at the ACM SIGEVO organized GECCO 2009 conference.

Dr. Xavier Llora interests and work on genetics-based machine learning (GBML) have earned him a spot among the leaders of the renaissance of learning classifier systems (LCSs). His 2002 PhD dissertation challenged traditional data-mining techniques by showing the effectiveness of GBML approaches. He has help organize two edition of the International Workshop of Learning Classifier Systems books and their proceedings. He served as LCS/GBML track chair in the ACM SIGEVO organized GECCO 2005 conference and also organized the NCSA/IlliGAL Gathereing on Evolutionary Learning in 2006. Llora, awarded with a young research fellowship by the Catalan Government, started his academic career at the Ramon Llull University (Barcelona, Spain) where he earned his PhD degree with honors. In 2003 he moved to the University of Illinois at Urbana-Champaign where he joined the Illinois Genetic Algorithms Laboratory. Since then, Llora has headed the DISCUS project (www-discus.ge.uiuc.edu), a project to support collaborative human-innovation and creativity. In summer 2005, Llora was named research assistant professor at the University of Illinois at Urbana-Champaign and in 2006 a NCSA affiliate where he pursues the development of data-intensive computing infrastructure in the cloud for large-scale data and text mining under the SEASR project (seasr.org).

Representation is a central issue in evolutionary computation. The no free lunch theorem proves that there is no intrinsic advantage in a particular algorithm when considered against complete spaces of problems. The corollary is that algorithms should be fitted to the problems they are solving. Choice of representation is the point, in the design of an evolutionary algorithm, where the designer has the greatest influence over the adaptive landscape and behavior of the algorithm. That suggests that a large library of representations should be a part of the tool-kit of any evolutionary computation researcher.

This survey will cover a broad variety of representations with comments on their domains of applicability and available variation operators. The representations covered include:

• Standard, linear, Cartesian and context free grammar genetic programming
• Genetic programming with GP-Automata
• The Sierpinski representation for parameter estimation
• Representations for searching spaces of networks
• Representations for game-playing agents including lookup tables, artificial neural nets, finite state machines, function stacks, Markov chains, and techniques for implementing long term memory
• Standard and random key representations for ordered-gene problems
• Hybrid representations

Treatment of individual topics will be brief but a bibliography with pointers into the literature will be included and presentation of representations will be example driven.

Dr. Ashlock currently has over 150 peer reviewed scientific publications, a substantial majority of which are in computational intelligence. He has a PhD in mathematics from the California Institute of Technology. He has help positions on the faculty of programs in bioinformatics, ecology and evolutionary biology, human-computer interface, pure and applied mathematics, and mathematics education. He is the author of an undergraduate text on evolutionary computation and holds the Bioinformatics Chair in the Department of Mathematics and Statistics at the University of Guelph. HDr. Ashlock currently serves as the Chair of the IEEE Bioinformatics and Bioengineering Technical Committee and is an Associate Editor for the IEEE Transactions on Evolutionary Computation, The IEEE/ACM Transactions on Bioinformatics and Computational Biology, The IEEE Transaction on Computational Intelligence and Artificial Intelligence in Games, and Biosystems. His primary research focus is on solving problems with evolutionary computation by inventing problem-specific representations.

Numerical optimisation problems are abundant in science and engineering. Evolutionary algorithms (EAs) are often the method of choice for such tasks that are not efficiently solved by approaches from the repertoire of mathematical optimisation strategies. Examples include discontinuous, non-differentiable, and noisy problems.

The aim of this tutorial is to provide researchers facing numerical optimisation tasks with a comprehensive overview of modern EAs for such problems. Algorithms covered include Evolution Strategies from the simple (1+1)-ES to modern Covariance Matrix Adaptation Evolution Strategies (CMA-ES), as well as Differential Evolution, Particle Swarm algorithms, and other techniques. Invariance properties and the algorithms' ability to cope with ill-conditioning and noise will be discussed. Strategies will be compared based on theoretical insights on simple test functions and on the results of large-scale benchmarking exercises conducted in recent years.

Dirk Arnold is an Associate Professor with the Faculty of Computer Science at Dalhousie University in Halifax, Nova Scotia. His research interests are in evolutionary computation, optimisation, and computer graphics and animation. Dr. Arnold is an Associate Editor of both the IEEE Transactions on Evolutionary Computation and the MIT Press's Evolutionary Computation journal.

Anthropologists have long recognized the importance of Culture as a symbolic entity that emerges from individual experiences, co-evolves with the population, and in turn influences individual choices. How and why different Cultural forms have evolved in some environments and not others have always attracted the interest of scholars. In this tutorial Cultural Algorithms are employed as a modeling framework in which to address the emergence of complex cultural structures and to study their impact on a population of individuals. It addresses both theoretical and practical issues in the development of large-scale multi-agent systems for engineering problem solving, simulation, and reality games. A recent reality game application of Cultural Algorithms was selected as one of the top 100 Scientific discoveries of 2009 by Discover Magazine.

Dr. Robert G. Reynolds received his Ph.D. degree in Computer Science, specializing in Artificial Intelligence, in 1979 from the University of Michigan, Ann Arbor. He is currently a professor of Computer Science and director of the Artificial Intelligence Laboratory at Wayne State University. He is an Adjunct Associate Research Scientist with the Museum of Anthropology at the University of Michigan-Ann Arbor. He is also affiliated with the Complex Systems Group at the University of Michigan-Ann Arbor and is a participant in the UM-WSU IGERT program on Incentive-Based Design. His interests are in the development of computational models of cultural evolution for use in the simulation of complex organizations and in computer gaming applications. Dr. Reynolds produced a framework, Cultural Algorithms, in which to express and computationally test various theories of social evolution using multi-agent simulation models. He has applied these techniques to problems concerning the origins of the state in the Valley of Oaxaca, Mexico, the emergence of prehistoric urban centers, the origins of language and culture, and the disappearance of the Ancient Anazazi in Southwestern Colorado using game programming techniques. He has co-authored three books; Flocks of the Wamani (1989, Academic Press), with Joyce Marcus and Kent V. Flannery; The Acquisition of Software Engineering Knowledge (2003, Academic Press), with George Cowan; and Excavations at San Jose Mogote 1: The Household Archaeology with Kent Flannery and Joyce Marcus (2005, Museum of Anthropology-University of Michigan Press).

He has received funding from both government and industry to support his work. He has published over 250 papers on the evolution of social intelligence in journals, book chapters, and conference proceedings. The journals include IEEE Computer, IEEE Computational Intelligence, Complexity, Scientific American, IEEE Transactions of Evolutionary Computation, IEEE Transactions on Systems, Man, and Cybernetics, IEEE Software, Communications of the ACM, and the Proceedings of the National Academy of Sciences. He is also co-editor of four books on evolutionary computation. Recently, a paper co-authored with M. Ali was selected as the Best Paper of 2008 in the International Journal of Intelligent Computing and Cybernetics.

Dr. Reynolds currently teaches courses on Cultural Algorithms, Computational Intelligence in Games, Artificial Intelligence, Social Intelligence, Evolutionary Computation, and Agent-Based Modeling. In addition, he has given a number of tutorials on Cultural Algorithms, including the IEEE 2007 Spring Symposium on Computational Intelligence.

He is currently an associate editor for the IEEE Transactions on Computational Intelligence in Games, IEEE Transactions on Evolutionary Computation, International Journal of Swarm Intelligence Research, International Journal of Artificial Intelligence Tools, International Journal of Computational and Mathematical Organization Theory, International Journal of Software Engineering and Knowledge Engineering, and the Journal of Semantic Computing. He was also a program co-chair for the 2008 IEEE World Congress on Computational Intelligence, program co-chair for 2008 IEEE Swarm Intelligence Symposium, on the Advisory Board for the International Swarm Intelligence Symposium (2007), and Plenary Chair for 2010 IEEE Swarm Intelligence Conference.

This tutorial elaborates on parameter tuning from three perspectives. First, we investigate possible definitions of evolutionary algorithm (EA) parameters and their impact on the very notion of EAs. Here we also (re)consider the "myth" of robust EA parameters. Second, we identify the core challenge of calibrating EA parameters and give an extensive overview of tuning methods, showing many examples. Third, we discuss the practical and theoretical consequences of good and fast parameter tuners on the evolutionary computing methodology of the (near) future. The tutorial is concluded with a number of concrete recommendations for EC researchers and practitioners.

A.E. Eiben is full professor on the Free University Amsterdam. He is head of the Computational Intelligence Group at the Computer Science Department He is one of the European early birds of Evolutionary Computing, his first EC paper dates back to 1989. Since then he has published over 150 research papers, co-edited and co-authored various volumes, including the best-seller “Introduction to Evolutionary Computing” (Springer, 2003, 2007). He has been organizing committee member of practically all major international evolutionary computing conferences (CEC, EP, EuroGP, EvoStar, FOGA, GECCO, PPSN). He is editorial board member of five international journals and series editor for Springer’s book series on Natural Computing. He is also member of numerous science management bodies, e.g., of the IEEE Computer Society Technical Committee on Computational Intelligence, Executive Board of the European Network on Evolutionary Computing, Steering Committee for the Parallel Problem Solving from Nature (PPSN) conference series. His recent research interests include tuning and control of evolutionary algorithm parameters, methodology, and on-line embodied evolution in robotics.

Many difficult real-life control design problems can be formulated in the framework of nonlinear optimal control theory. Dynamic programming formulation offers the most comprehensive solution approach to nonlinear optimal control in a state feedback form, which is desirable because of its beneficial properties like online applicability (because of a closed-form solution), robustness with respect to noise and modeling uncertainties. However, solving the associated Hamilton-Jacobi-Bellman (HJB) equation demands a very large (rather infeasible) amount of computations and storage space dedicated for this purpose (popularly known as “curse-of-dimensionality” issue).

An innovative idea is to get around this numerical complexity by using an “Approximate Dynamic Programming (ADP)” formulation, the solution of which is obtained through a dual neural network approach called Adaptive Critic (AC) design. The AC method determines an optimal control law for a system by successively adapting two neural networks, an action network (which dispenses the control signals) and a critic network (which ‘learns’ the desired performance index for some function associated with the performance index). This adaptive critic optimal control synthesis approach has many desirable features, viz. having a feedback form of the control, ability for on-line implementation, no need for approximating the nonlinear system dynamics with linearization or quasi-linearization etc.

The adaptive critic design has been successfully demonstrated in a large number of challenging application problems, including power systems control and optimization, process control, sensors and sensor networks, and biomedical control.

The objective of this tutorial is to provide the audience a brief exposure to the philosophy of adaptive critic design, especially to give them a good exposure to a wide variety of applications. Application problems to be presented in this tutorial will include various topics as mentioned above, which will be derived largely from literature and the research experience of the speaker in this field.

Ganesh Kumar Venayagamoorthy received a Ph.D. degree in electrical engineering from the University of KwaZulu Natal, Durban, South Africa in 2002. Currently, he is an Associate Professor of Electrical and Computer Engineering, and the founder and the Director of the Real-Time Power and Intelligent Systems (RTPIS) Laboratory at Missouri University of Science and Technology (Missouri S&T). He was a Visiting Researcher with ABB Corporate Research, Sweden, in 2007. His research interests are in the development and applications of advanced computational algorithms for real-world applications, including power systems stability and control, smart grid applications, sensor networks and signal processing. He has published two edited books, five book chapters, and over 80 refereed journals papers and 270 refereed conference proceeding papers. He has been involved in approximately US$ 7 Million of competitive research funding. Dr. Venayagamoorthy is a Fellow of the Institution of Engineering and Technology (IET), UK and the South African Institute of Electrical Engineers. He is a Senior Member of the IEEE and the International Neural Network Society, and a Member of the American Society for Engineering Education. He is member of Board of Governors of INNS.

Due to the computational and theoretical advantages that processing in the complex domain offers over real valued bivariate vectors, the area of complex-valued neural networks is one of fastest growing research areas in the neural network community. In addition, recent progress in pattern recognition, robotics, radar and sonar, mathematical biosciences, and renewable energy, has brought to light problems where nonlinearity, multidimensional data natures, uncertainty, and complexity play major roles – complex-valued neural networks are a natural computational model to account for these classes of applications.

The most important notion underlying the theory of complex-valued neural networks is that of the phase information. This enables us to employ advanced concepts, such as phase synchrony and coherence, and to model simultaneously the amplitude-phase relationships for a range of computational scenarios, including spectrum estimation, self-organizing maps, classification, adaptive filtering, and pattern recognition. Apart from standard applications where signals are complex ‘by design’ (communications, spectrum estimation), complex domain processing has proved extremely beneficial for problems which are cast into the complex domain by ‘convenience of representation’. This includes modeling of directional processes, where the amplitude and direction (phase) are naturally combined into a complex-valued model, such as in wind forecasting, direction of arrival estimation, radar, sonar, and sensor array processing.

Owing to recent advances in the statistics of complex variable, complex-valued neural network models have been shown not only to exhibit enhanced accuracy, but also to facilitate physical interpretation of their variables. The synergy of complex nonlinearity, noncircularity of probability distributions, nonlinear separability, self-organization, and the supporting optimization theories underpins this tutorial, which aims at providing a rigorous unifying framework for the design, analysis, and interpretation of complex neural network models, and a convenient platform for numerous practical applications. The material is supported by detailed case studies across engineering disciplines, highlighting the practical usefulness of complex-valued neural networks.

The tutorial consists of three parts. Part I will be presented by Dr Mandic and will cover recent advances in complex statistics, complex noncircularity, and temporal complex neural networks models. In Part II, Dr Aizenberg will introduce neurons with phase-dependent activation functions and will illustrate their usefulness in learning non-linearly separable problems and solving multiple-class classification problems. Dr Hirose will present Part III, and will focus on self-organisation and recent working solutions and practical applications of complex-valued neural networks in microwave- and millimetre-wave radar systems.

The material will be of interest to a wide audience, spanning the areas of adaptive signal processing, machine learning, image processing, object recognition, radar, and mathematical biosciences.

Recurrent neural networks offer a framework for the identification of dynamical systems. Based on a so called correspondence principle of neural equations and architectures we will discuss the identification of open as well as closed systems. For these general frameworks we have to discuss scalability, mixture of time scales of the dynamics, the learning dependent on large/small datasets, and more.

Applying this framework to forecasting additional questions show up: what is the correct framework for long term forecasting, how to handle uncertainty and risk in forecasting.

Finally we will apply the above fundamentals to control: we will work out an integrated approach for the system identification and open loop, as well as closed loop control laws.

The workshop is of interest for academics as well as practitioners. It gives an overview of 22 years of our method and software development together with econometric applications we have done at Siemens Corporate Technology.

Methods of support vector machines and kernel-based learning have been successful on a wide range of applications especially for problems with high dimensional inputs. Different methodologies have emerged with use of optimization-based settings, estimation in reproducing kernel Hilbert spaces and probabilistic approaches.

In this tutorial we outline new directions in unsupervised learning, with emphasis on models that possess primal and dual model representations, using feature maps and positive definite kernels, respectively. Benefits will be shown with respect to out-of-sample extensions, new model selection procedures, the ability to incorporate prior knowledge and handling large data sets.

Topics include:

• Kernel maps for dimensionality reduction and data visualization: eigenmap methods, introduction of reference point, linear system solving, learning and generalization
• Kernel component analysis: kernel principal component analysis, weighting, sparse and robust versions
• Kernel canonical correlation analysis as a measure for independence
• Kernel spectral clustering: model selection, out-of-sample extensions, prior knowledge incorporation, highly sparse representations, large data sets
• Application examples include power grid networks, time-series analysis, image segmentation, literature mining

Traditional computational models of neural networks are based on models of neurons that produced only an output amplitude, and not on the precise timing of these outputs. Recent research in the field of neuroscience  has indicated the tremendous importance of the role of timing in real networks of neurons present in vertebrate brains.

The purpose of this tutorial is to bridge the gap between current neuroscientific observations and the computational modeling of timing information in neural networks. Our goal is to develop deeper understanding of how the time domain can be effectively employed in neural network models. This will be achieved through the use of oscillators that model temporal firing patterns of a neuron or a group of neurons. There is a growing body of research on the use of oscillatory neural networks, and their desirable propertie. A network of oscillatory elements can exhibit synchronization under the right conditions. Such networks of synchronizing elements have been shown to be effective in solving the binding problem, which is of great significance in the field of neuroscience.

The oscillatory neural models can be employed at multiple scales of abstraction, ranging from individual neurons, to groups of neurons using Wilson-Cowan modeling techniques and eventually to the behavior of entire brain regions as revealed in oscillations observed in EEG recordings. Hence this topic is of emerging interest in the brain sciences, as imaging techniques are able to resolve sufficient temporal detail to provide an insight into how the time domain is deployed in cognitive function. Furthermore, new optical stimulation techniques enable a new method of testing predictions that can be made through computational modeling, which pave the way for a tighter integration between the experimentation and theory.

Semi-supervised learning is an active topic in both the research and application fields of data mining. In many applications, labeled data are usually expensive to obtain and unlabeled data are widely observed. Semi-supervised learning is important in that the unlabeled data can help to improve the performance of supervised learning and thus greatly reduces the human effort in labeling data. In this tutorial, we will first introduce the fundamental assumptions in semi-supervised learning. Based on these assumptions, we will introduce the related algorithms, including self-training, co-training, EM-based methods, graph-based methods, and large-margin based methods. To better understand these algorithms, we will show the demos. Furthermore, we will also introduce some applications of these algorithms. In particular, we will present a study of these semi-supervised learning algorithms in privacy preservation in social network analysis. Finally, we will review recent advances and future perspectives in semi-supervised learning.

This tutorial will set out to address the key limiters of scalability and discuss the means of increasing the numbers of devices on a chip to biologically plausible levels, i.e., investigate biologically inspired connectivity solutions for future nano-electronic systems.

The significance of brain-inspired connectivity comes from the fact that the mammalian brain is one of the most efficient and remarkably reliable network of processing elements currently know to mankind. One reason that scalability is so important is that much of the brain’s computing power comes from its massive parallelism.

This tutorial will analyze the global connectivity of the brain along with the detail ways it communicates locally. We will first of all compare the brain’s connectivity (based on neurological data) with well-known computer network topologies (originally used in super-computers). The comparison will reveal that brain’s connectivity is in good agreement with Rent’s rule. However, the known network topologies fall short of being strong contenders for mimicking the brain, therefore emphasis will be placed upon detailed Rent-based (top-down) modeling of generic two-layer hierarchical networks. This analysis will identify those generic two-layer hierarchical network topologies which when combined could mimic brain’s connectivity. The range of granularities (i.e., number of gates/cores/neurons) where such mimicking is possible will be presented and discussed. Accurate wire length estimates for hierarchical networks with complexity tending to that of the brain will follow, and will help in estimating many other important parameters like power, energy, and reliability.

For local interconnects, artificial synapses as well as axonal communication will be evaluated. Issues in terms of latency, energy dissipated and signal integrity are inherent problems that normally act to limit the scalability and can negate any computational advantages of parallelism. Various schemes used for reducing the interconnecting density such as Pulsed Wave Interconnect, Address Event Decoding, and Multiple Valued Logic, all have deficiencies which prevent them from scaling towards biologically plausible levels, while Time Multiplexed Architectures seems to exhibit stronger potential. Finally, very fresh axonal-inspired communications will be introduced, and their performance will be thoroughly evaluated.

These results should have immediate implications for the design of future networks-on-chip (NoCs) in general (in the short term), and for the burgeoning field of multi-/many-core processors in particular (in the medium term), as well as for forward-looking investigations on emerging (brain-inspired) nano-architectures (in the long run).

Data mining systems contain a large (and quickly growing) number of machine learning methods based on neural, fuzzy, pattern recognition and statistical ideas. Despite significant progress in various theoretical and applied areas many problems remain unsolved, comprehensive theory presenting a unified perspective on various learning methods is missing, large component-based data mining packages contain now hundreds of learning methods, input transformations, pre- and post-processing components that may be combined in more than 10 million ways. Although there is "no free lunch" (no single method is the best for all test problems) several methods that are close to optimal may be found through meta-learning based on heuristic search in the space of all possible learning models. Various model spaces are considered as the basis for meta-learning: 1) similarity-based algorithms that identify prototypes and optimize similarity measures; 2) heterogeneous systems, that include neural, fuzzy, prototype-based and hierarchical partitioning algorithms; 3) general transformation-based systems.

Most general implementation of meta-learning is possible within transformation-based learning paradigm that unifies most of computational intelligence research and shows how to solve the "crises of the richness" selecting optimal transformations to minimize complexity and maximize quality of the resulting data models. Meta-learning systems learn simplest data models that many sophisticated methods miss, generate multi-resolution models whenever needed, and solve difficult, highly non-separable problems that are beyond capabilities of current state-of-the-art algorithms, including neural networks and support vector machines. In contrast to backpropagation that tries to achieve linear separability in one shot additional criteria are defined after each transformation to create appropriate internal representations. Visualization of learning dynamics in transformation-based systems shows how to set simpler goals for learning, for example k-separability instead of linear separability.

This tutorial will include:

1. Review of various approaches to meta-learning.
2. Meta-learning as search in the model space - general idea.
3. Model space based on similarity-based learning.
4. Model space based on composition of transformations creating internal representations.
5. Complexity control of the search process.
6. Some implementation details.
7. Examples and some lessons learned from the use of meta-learning.

Project page: http://www.is.umk.pl/projects/meta.html.

This tutorial introduces molecular biology for computational scientists that are tangentially familiar with or unfamiliar with it. The tutorial will cover molecular biology with an emphasis on data types, data sources, and potential applications for computational intelligence. Biological topics covered include the central dogma of molecular biology, issues connected with protein structure and folding, the problems concerning RNA folding, motif finding, sequence comparison techniques including BLAST, dynamic programming, string kernels, and more exotic techniques, and microarray technology and data sets including comments on clustering methods and dimension reduction. The tutorial will include pointers to both biological data sets and the CI literature on bioinformatics. A series of examples of applications of computational intelligence techniques with both the representational structure and application domain will form a part of the presentation.

Dr. Ashlock currently has over 150 peer reviewed scientific publications, a substantial majority of which are in computational intelligence. He has a PhD in mathematics from the California Institute of Technology. He has help positions on the faculty of programs in bioinformatics, ecology and evolutionary biology, human-computer interface, pure and applied mathematics, and mathematics education. He is the author of an undergraduate text on evolutionary computation and holds the Bioinformatics Chair in the Department of Mathematics and Statistics at the University of Guelph. HDr. Ashlock currently serves as the Chair of the IEEE Bioinformatics and Bioengineering Technical Committee and is an Associate Editor for the IEEE Transactions on Evolutionary Computation, The IEEE/ACM Transactions on Bioinformatics and Computational Biology, The IEEE Transaction on Computational Intelligence and Artificial Intelligence in Games, and Biosystems. His primary research focus is on solving problems with evolutionary computation by inventing problem-specific representations.

Dr. Stefan C. Kremer is an associate professor in the Department of Computing and Information Science at the University of Guelph and Director of the Bioinformatics Graduate Program. His research focus is on structural pattern induction--that is, the identification, recognition and application of underlying patterns and rules in structured data including strings, languages, time-series, 3-dimensional objects, etc.. His most recent work focusses on biological data including DNA, RNA and protein sequences.

Reactive Search Optimization (RSO) advocates the integration of sub-symbolic machine learning techniques into heuristics for solving complex optimization problems. The word reactive hints at a ready response to events during the problem solving process through an internal online feedback loop for the self-tuning of critical parameters. Methodologies of interest for RSO include machine learning and statistics, in particular reinforcement learning, active or query learning, neural networks, computational intelligence.

Intelligent optimization, a superset of Reactive Search Optimization, refers to a more extended area of research, including online and offline schemes based on the use of memory, adaptation, incremental development of models, experimental algorithmics applied to optimization, intelligent tuning and design of heuristics. The tutorial focusses on the main methods and corresponding software tools.

Prof. Roberto Battiti received the Laurea degree in Physics from the University of Trento, Italy, in 1985 and the Ph.D. degree from the California Institute of Technology (Caltech), USA, in 1990. He is now full professor of Computer Science at Trento university, deputy director of the DISI Department (Electrical Engineering and Computer Science) and director of the LION lab at (machine Learning and Intelligent OptimizatioN). His main research interests are heuristic algorithms for optimization problems, in particular reactive search optimization algorithms for discrete optimization problems. R. Battiti is a fellow of the IEEE. Full details about interests, research activities and scientific production can be found in the web: lion.dit.unitn.it/~battiti , reactive-search.org.

Clustering, also known as unsupervised learning or exploratory data analysis, attempts to explore the unknown natures of data through the separation of a finite dataset, with little or no ground truth, into a finite and discrete set of “natural,” hidden data structures. Clustering has long been known as one of the most important and primitive activities of human beings, with rich theories and applications arising from a wide variety of communities, ranging from engineering and computer sciences (computational intelligence, data mining, information retrieval, machine learning, pattern recognition, mechanical engineering, electrical engineering), life and medical sciences (biology, clinic, genetics, microbiology, paleontology, pathology, psychiatry, phylogeny), and astronomy and earth sciences (geography, geology, remote sensing), to social sciences (anthropology, archeology, education, psychology, sociology) and economics (business, marketing). Such diversity, while providing many options, inevitably causes confusion because of the differing terminologies and goals and the lack of good communication between these communities. As such, the goal of the tutorial is to have the audience understand the status quo of this field, thus avoiding unnecessary burdens and saving development time.

The presentation will begin with an introduction and discussion of the basic and major problems relating to cluster analysis. While we will provide a review of the important, influential, and state-of-the-art clustering algorithms in the literature, we will concentrate on the clustering algorithms rooted in neural networks, evolutionary computation, and fuzzy set theory. We will also illustrate the applications of this clustering in situations such as the traveling salesman problem, bioinformatics, situation awareness, and so on. The tutorial will conclude with an analysis of the perspectives and challenges of clustering.

The outline of the tutorial is summarized as follows:

1. Clustering basics
   • Pattern classification and clustering
   • What is a cluster?
   • Basic procedure of cluster analysis
   • Clustering is everywhere
   • Proximity measures
   • Cluster validity
2. Clustering algorithms
   • Hierarchical clustering: Agglomerative and divisive
   • Squared error-based clustering: K-means and its variants
   • Mixture density-based clustering
   • Kernel-based clustering: Support vector clustering and the nonlinear version of the existing linear clustering algorithms
   • Large-scale data clustering: Incremental clustering, divide-and-conquer, random sampling methods
   • Fuzzy clustering: Fuzzy c-means and its variants
   • Evolutionary computational technology-based clustering: Particle swarm optimization, differential evolution, ant algorithm, genetic algorithm, and/or hybrid of meta-heuristics
   • Neural network-based clustering: Hard competitive learning clustering and soft competitive learning clustering
3. Clustering applications
   • Traveling salesman problem
   • Bioinformatics
   • Situation awareness
   • Mineral identification
4. Clustering trends