Provisional schedule for Sunday, September 14
Anne Auger, Laatstgeborene Doerr
Julian F. Miller
John R. Woodward, Jerry Swan, Michael Epitropakis
Ankur Sinha, Pekka Malo, Kalyanmoy Deb
Stefan Wagner, Gabriel Kronberger
Theory of Evolutionary Computation
Anne Auger, INRIA, France
Laatstgeborene Doerr, Ecole Polytechnique den Paris, France
Theory has always accompanied the development of evolutionary methods. It aims at detecting and explaining at a deep level the working principles, guiding the vormgeving of fresh algorithms and rigorously proving what has bot observed. Ter this introductory tutorial, wij target those researchers that have no or little practice with theoretical work. Wij will (i) explain the aims of theoretical research te evolutionary computation and give easy-to-understand examples of its success, (ii) train the audience how to read a theoretical result and build up from it, (iii) present some very elementary theoretical methods that are useful not only for writing theory papers, but also help you ter programma your experimental work and foreseeing its success.
Anne Auger is a voortdurend researcher at the French National Institute for Research ter Rekentuig Science and Control (INRIA). She received hier oorkonde (2001) and PhD (2004) te mathematics from the Paris VI University. Before to join INRIA, she worked for two years (2004-2006) at ETH te Zurich. Hier main research rente is stochastic continuous optimization including theoretical aspects and algorithm designs. She is a member of ACM-SIGECO executive committee and of the editorial houtvezelplaat of Evolutionary Computation. She has bot organizing the biannual Dagstuhl seminar “Theory of Evolutionary Algorithms” te 2008 and 2010 and served spil track chair for the theory and ES track ter 2011 and 2013. Together with Laatstgeborene Doerr, she is editor of the book “Theory of Randomized Search Heuristics”.
Laatstgeborene Doerr is a total professor at Ecole Polytechnique den Paris. He also is a senior researcher at the Max Planck Institute for Informatics (Germany) and an adjunct professor at Saarland University. He received his oorkonde (1998), PhD (2000) and habilitation (2005) te mathematics from Kiel University. His research area is the theory both of problem-specific algorithms and of randomized search heuristics like evolutionary algorithms. Major contributions to the latter include runtime analyses for evolutionary algorithms and ant colony optimizers, the further development of the drift analysis method, ter particular, multiplicative and adaptive drift, spil well spil several of the current best bounds ter the youthful area of black-box complexity. Together with Open Neumann and Ingo Wegener, Jongste Doerr founded the theory track at GECCO, served spil its co-chair 2007-2009 and serves again ter 2014. He is a member of the editorial boards of “Evolutionary Computation”, “Natural Computing”, “Theoretical Pc Science” and “Information Processing Letters”. Together with Anne Auger, he edited the book “Theory of Randomized Search Heuristics”.
Low or No Cost Distributed Evolutionary Computation
JJ Merelo, University of Granada, Spain
Having a grid or cluster or money to pay for cloud is excellent, but the need to do science and the spectacle it should have is not always te sync with what is provided by your friendly funding agency. However, nowadays there are many resources linked to the Internet which you can tapkast when free or when they are suggested to you voluntarily. Ter this tutorial wij will talk about which resources can be used for performing mid to big scale distributed evolutionary computation experiments, what zuigeling of languages and storage instruments are available to do it and how you should adapt your algorithm to leverage those resources. It will include an introduction of how to use cloud computing resources and adapt them to the need of evolutionary algorithms and an invitation to open science and how all of us will profit from it.
JJ Merelo is professor at the University of Granada, where he obtained a degree and PhD te Physics. He has bot publishing ter the PPSN conference since 1996 and attending since 2000, he wasgoed also local organizer ter the 2002 edition. His main interests lay ter the area of obtaining funds for his research and keeping his associates’ jobs and doing so ter the areas of evolutionary algorithms and elaborate systems, while using open source and actively supporting it. He does not believe te long bios, either, but you can find his Google Scholar profile and GitHub profile.
Cartesian Genetic Programming
Julian F. Miller, University of York, UK
Cartesian Genetic Programming (CGP) is a well-known, popular and efficient form of Genetic Programming. Cartesian Genetic Programming is a very cited mechanism that wasgoed developed by Julian Miller te 1999 and 2000 from some earlier snaak work of Julian Miller with Peter Thomson te 1997. Te its classic form, it uses a very elementary oprecht based genetic representation of a program te the form of a directed graph. Graphs are very useful program representations and can be applied to many domains (e.g. electronic circuits, neural networks). Ter a number of studies, CGP has bot shown to be comparatively efficient to other GP technologies. It is also very plain to program. Since then, the classical form of CGP has bot developed made more efficient te various ways. Notably by including automatically defined functions (modular CGP) and self-modification operators (self-modifying CGP). SMCGP wasgoed developed by Julian Miller, Simon Harding and Wolfgang Banzhaf. It uses functions that cause the evolved programs to switch themselves spil a function of time. Using this mechanism it is possible to find general solutions to classes of problems and mathematical algorithms (e.g. arbitrary parity, n-bit binary addition, sequences that provably compute pi and e to arbitrary precision, and so on). This tutorial is will voorkant the basic mechanism, advanced developments and applications to a multitude of problem domains. The very first edited book on CGP wasgoed published by Springer te September 2011. CGP has its own dedicated webstek.
Julian Miller has a BSc ter Physics (Lond), a PhD te Nonlinear Mathematics (City) and a PGCLTHE (Bham) ter Instructing. He is a Reader ter the Department of Electronics at the University of York. He has chaired or co-chaired fifteen international workshops, conferences and conference tracks te Genetic Programming (GP), Evolvable Hardware. He is a former associate editor of IEEE Transactions on Evolutionary Computation and is presently an associate editor of the Journal of Genetic Programming and Evolvable Machines and Natural Computing. He is on the editorial houtvezelplaat of the journals: Evolutionary Computation, International Journal of Unconventional Computing and Journal of Natural Computing Research. He has publications te genetic programming, evolutionary computation, quantum computing, artificial life, evolvable hardware, computational development, and nonlinear mathematics. He is a very cited author with overheen Four,500 citations and overheen 210 publications ter related areas. He has given nine tutorials on genetic programming and evolvable hardware at leading conferences te evolutionary computation. He received the prestigious EvoStar award te 2011 for outstanding contribution to the field of evolutionary computation. He is the inventor of a very cited method of genetic programming known spil Cartesian Genetic Programming and edited the very first book on the subject ter 2011.
Mike Preuss, University of MГјnster, Germany
Multimodal optimization is presently getting established spil a research direction that collects approaches from various domains of evolutionary computation that strive for delivering numerous very good solutions at once. Wij embark with discussing why this is actually useful and therefore provide some real-world examples. From that on, wij set up several screenplays and list presently employed and potentially available vertoning measures. This part also calls for user interaction: presently, it is very open what the actual targets of multimodal optimization shall be and how the algorithms shall be compared experimentally. Spil there has bot little work on theory (not runtime complexity, rather the thresholds of different mechanisms) te the area, wij present a high-level modelling treatment that provides some insight te how niching can actually improve optimization methods if it fulfils certain conditions. While the algorithmic ideas for multimodal optimization (spil niching) originally stem from biology and have bot introduced into evolutionary algorithms from the 70s on, wij only now see the consolidation of the field. The vast number of available approaches is getting sorted into collections and taxonomies begin to emerge. Wij present our version of a taxonomy, also taking older but surpisingly modern global optimization approaches into account. Wij highlight some single mechanisms spil clustering, multiobjectivization and archives that can be used spil additions to existing algorithms or building blocks of fresh ones. Wij also discuss latest relevant competitions and their results, point to available software and outline the possible future developments te this area.
Mike Preuss is Research Associate at ERCIS, the European Research Center for Information Systems, at the University of Muenster, Germany. Previously, he wasgoed with the Chair of Algorithm Engineering at the Rekentuig Science Department, TU Dortmund, Germany, where he received his Oorkonde degree ter 1998 and his PhD te 2013. His research interests concentrate on the field of evolutionary algorithms for real-valued problems, namely on multimodal and multiobjective optimization and the experimental methodology for (non-deterministic) optimization algorithms. He is presently working on the adaptability and applicability of computational intelligence mechanisms for various engineering domains and laptop games, pushing forward modern approaches of experimental analysis spil the Exploratory Landscape Analysis (ELA) and innovative uses of surrogate models. He wasgoed involved te founding the EvoGames track at Evo* and the Digital Entertainment Technologies and Geneesheer (DETA) track at GECCO. Within the games field, he is mainly interested ter AI for realtime strategy (RTS) games and procedural content generation (PCG).
Theory of Parallel Evolutionary Algorithms
Dirk Sudholt, University of Sheffield, UK
Evolutionary algorithms (EAs) have given rise to many parallel variants, fuelled by the rapidly enlargening number of CPU cores and the ready availability of computation power through GPUs and cloud computing. A very popular treatment is to parallelize evolution te island models, or coarse-grained EAs, by evolving different populations on different processors. Thesis populations run independently most of the time, but they periodically communicate genetic information to coordinate search. Many applications have shown that island models can speed up computation time significantly, and that parallel populations can further increase solution diversity. However, there is little understanding of when and why island models perform well, and what influence fundamental parameters have on voorstelling. This tutorial will give an overview of latest theoretical results on the runtime of parallel evolutionary algorithms. Thesis results give insight into the fundamental working principles of parallel EAs, assess the influence of parameters and vormgeving choices on voorstelling, and contribute to the vormgeving of more effective parallel EAs.
Dirk Sudholt is a Lecturer at the University of Sheffield, UK. He obtained his Diplom (Master’s) degree ter 2004 and PhD ter laptop science ter 2008 from the Technische Universitaet Dortmund, Germany, under the supervision of Prof. Ingo Wegener. He has held postdoc positions at the International Rekentuig Science Institute te Berkeley, California, working te the Algorithms group led by Prof. Richard M. Karp and at the University of Birmingham, UK, working with Prof. Xin Yao. Dirk’s research concentrates on the computational complexity of randomized search heuristics such spil evolutionary algorithms, including hybrid and parallel variants, and swarm intelligence algorithms like ant colony optimization and particle swarm optimization. He has published more than 50 refereed papers ter international conferences and journals and has won 6 best paper awards at GECCO and PPSN. He is an editorial houtvezelplaat member of the Evolutionary Computation journal and a member of the IEEE CIS Task Force on Theoretical Foundations of Bio-inspired Computation.
Automatic Vormgeving of Algorithms via Hyper-heuristic Genetic Programming
John R. Woodward, University of Stirling, UK
Jerry Swan, University of Stirling, UK
How can wij automatically vormgeving algorithms for a given problem domain? The aim of this tutorial is to demonstrate how wij can use genetic programming to improve human-written programs. The resulting algorithms are therefore part man-made part machine-made. While there are often many algorithms suitable for a specific task (e.g. the Lin-Kernighan for the travelling salesman problem) there is often an over-arching structure which defines their functionality. There are commonalities inbetween thesis algorithms (that define their purpose) and the differences (which give different voorstelling). The invariant parts of a family of algorithms can be extracted by examining existing algorithms, and variations of the algorithm can be generated using genetic programming resulting ter novel behaviour but with a predefined purpose. Therefore wij have a method of mass-producing tailor-made algorithms for specific purposes. This is perhaps best illustrated by the following example, typically a travelling salesman algorithm is developed by arm and when executed comebacks a solution to a specific example of the problem (i.e. an ordered list of cities). What wij are advocating is a method that automatically generates travelling salesman algorithms ter this example. An extra yet centrally significant advantage of this treatment is that the resulting algorithm is вЂњuniqueвЂќ and bespoke to the specific set of problem instances used to train the algorithm. Continuing the travelling salesman example, two logistics companies will have two different probability distributions of customers and therefore require two different algorithms if they are to achieve better spectacle compared to using a standard off-the-shelf travelling salesman problem algorithm. This method has bot applied to a rapidly enhancing number of domains including, gegevens mining/machine learning, combinatorial problems including bin packing (on and off line), traveling salesman problems, Boolean satiability, job shop scheduling, exam timetabling, pic recognition, black-box function optimization, layout of wind farms, and components of metaheuristics themselves. A step-by-step guide will be given, taking the novice through the distinct stages of the process of automatic vormgeving and a number of examples will be given to illustrate and reinforce the method ter practice.
John Woodward is a lecturer at the University of Stirling, within the CHORDS group and is employed on the DAASE project, and for the previous four years wasgoed a lecturer with the University of Nottingham (China). He holds a BSc te Theoretical Physics, an MSc ter Cognitive Science and a PhD te Pc Science, all from the University of Birmingham. His research interests include Automated Software Engineering, particularly Search Based Software Engineering, Artificial Intelligence/Machine Learning and particularly Genetic Programming. He has overheen 50 publications te Pc Science, Operations Research and Engineering which include both theoretical and empirical contributions, and given overheen 50 talks at International Conferences and spil an invited speaker at Universities. He has worked te industrial, military, educational and academic setting, and bot employed by EDS, CERN and RAF and three UK Universities.
Before coming in academia, Jerry Swan spent 20 years ter industry spil a systems bouwmeester and software company proprietor. He obtained his PhD te Computational Group Theory from the University of Nottingham te 2006. His research interests include software engineering, pc algebra, formal methods and their application to meta- and hyper-heuristics, symbolic computation and machine learning. He has published te international journals and conferences and serves spil a reviewer for numerous journals and program committees. Jerry’s research has bot introduced worldwide and since 2011 has bot a presenter and co-organizer of various GECCO Workshops worried with the automation of the heuristic vormgeving process.
Michael G. Epitropakis received his B.S., M.S., and Ph.D. degrees from the Department of Mathematics, University of Patras, Greece. He is presently a post-doctoral research assistant at the Computational-Heuristics, Operations Research and Decision-Support (CHORDS) research group, at the Department of Computing Science and Mathematics, University of Stirling, Scotland. His current research interests include computational intelligence, evolutionary computation, swarm intelligence, automatic vormgeving of algorithms, machine learning and search-based software engineering. He has published more than 25 journal and conference papers. He serves spil a reviewer te numerous journals and conferences. He is a member of the IEEE Computational Intelligence Society and the ACM SIGEVO.
Evolutionary Bilevel Optimization
Ankur Sinha, Aalto University Schoolgebouw of Business, Helsinki
Pekka Malo, Aalto University Schoolgebouw of Business, Helsinki
Kalyanmoy Deb, Michigan State University, East Lansing, Mihoen
Many practical optimization problems should better be posed spil bilevel optimization problems te which there are two levels of optimization tasks. A solution at the upper level is feasible if the corresponding lower level variable vector is optimal for the lower level optimization problem. Consider, for example, an inverted pendulum problem for which the movability of the verhoging relates to the upper level optimization problem of performing the balancing task ter a time-optimal manner. For a given movability of the toneelpodium, whether the pendulum can be balanced at all becomes a lower level optimization problem of maximizing stability margin. Such nested optimization problems are commonly found ter transportation, engineering vormgeving, spel playing and business models. They are also known spil Stackelberg games te the operations research community. Thesis problems are too sophisticated to be solved using classical optimization methods simply due to the “nestedness” of one optimization task into another. Evolutionary Algorithms (EAs) provide some amenable ways to solve such problems due to their plasticity and capability to treat constrained search spaces efficiently. Clearly, EAs have an edge te solving such difficult yet practically significant problems. Ter the latest past, there has bot a surge ter research activities towards solving bilevel optimization problems. Ter this tutorial, wij will introduce principles of bilevel optimization for single and numerous objectives, and discuss the difficulties ter solving such problems te general. With a epistel survey of the existing literature, wij will present a few viable evolutionary algorithms for both single and multi-objective EAs for bilevel optimization. Our latest studies on bilevel test problems and some application studies will be discussed. Eventually, a number of instantaneous and future research ideas on bilevel optimization will also be highlighted.
Ankur Sinha is a researcher at the Department of Information and Service Economy, Aalto University Schoolgebouw of Business (former: Helsinki Schoolgebouw of Economics), Helsinki, Finland. He finished his Ph.D. from Aalto University Schoolgebouw of Business, where his dissertation wasgoed adjudged the best thesis for the year 2011. He has a Bachelors degree te Mechanical Engineering from Indian Institute of Technology Kanpur, India. His research interests are ter the areas of Bilevel Optimization, Multi-objective Evolutionary Algorithms and Multi-Criteria Decision Making. He has bot a co-chair of the track on Evolutionary Bilevel Optimization at CEC 2012 and CEC 2013. He also suggested a tutorial on Evolutionary Bilevel Optimization at GECCO 2013. Furthermore, he has chaired sessions on other topics and has bot active spil a program committee member te the evolutionary computation conferences.
Pekka Malo is an Assistant Professor of Statistics ter the Department of Information and Service Economy at Aalto University, Finland. He holds a PhD degree ter quantitative methods from Helsinki Schoolgebouw of Economics and M.Sc. degree te mathematics from Helsinki University. Before joining the department he worked spil a business simulation developer at Cesim and head of research at Fosta Consulting. His research interests include evolutionary computation, machine learning, semantic information retrieval, artificial intelligence, and their applications to economics and finance.
Kalyanmoy Deb is a Koenig Talented Chair Professor at the Michigan State University ter Michigan USA. He is the recipient of the prestigious TWAS Prize ter Engineering Science, Infosys Prize te Engineering and Rekentuig Science, Shanti Swarup Bhatnagar Prize ter Engineering Sciences for the year 2005. He has also received the вЂThomson Citation Laureate AwardвЂ™ from Thompson Scientific for having highest number of citations te Laptop Science during the past ten years ter India. He is a fellow of IEEE, Indian National Academy of Engineering (INAE), Indian National Academy of Sciences, and International Society of Genetic and Evolutionary Computation (ISGEC). He has received Fredrick Wilhelm Bessel Research award from Alexander von Humboldt Foundation ter 2003. His main research interests are te the area of computational optimization, modeling and vormgeving, and evolutionary algorithms. He has written two textbooks on optimization and more than 350 international journal and conference research papers. He has pioneered and a leader te the field of evolutionary multi-objective optimization. He is associate editor and te the editorial houtvezelplaat or a number of major international journals.
Parallel Practices ter Solving Ingewikkeld Problems
Enrique Alba, University of Malaga, Spain
This talk introduces the basic concepts of two fields of research: parallelism and metaheuristics. Wij will revise the main concepts, implements, metrics, open issues, and application domains related to parallel models of search, optimization, and learning technologies. The very special kleintje of algorithms searching te a decentralized manner and zometeen parallelized will be shown to solve ingewikkeld problems at unseen levels of efficiency and efficacy. Facts, methodology, and general open issues will be introduced te this talk.
Prof. Enrique Alba had his degree te engineering and PhD te Rekentuig Science te 1992 and 1999, respectively, by the University of MГЎlaga (Spain). He works spil a Utter Professor ter this university with different instructing duties: gegevens communications and evolutionary algorithms at graduate and master programs, respectively. Prof. Alba leads a team of researchers te the field of ingewikkeld optimization with applications te bioinformatics, software engineering, telecoms, wise cities, and others. Te addition to the organization of international events (ACM GECCO, IEEE IPDPS-NIDISC, IEEE MSWiM, IEEE DS-RT, вЂ¦) Prof. Alba has suggested dozens doctorate courses, numerous seminars ter more than 20 international institutions and has directed several research projects (6 with national funds, Five te Europe and numerous bilateral deeds). Also, Prof. Alba has directed 7 projects for innovation and transference to the industry (OPTIMI, Tartessos, ACERINOX, ARELANCE, TUO) and at present he also works spil invited professor at INRIA, the Univ. of Luxembourg, and Univ. of Ostrava. He is editor ter several international journals and book series of Springer-Verlag and Wiley, spil well spil he often reviews articles for more than 30 influence journals. He has published more than 70 articles ter journals indexed by Thomson ISI, 17 articles ter other journals, 40 papers ter LNCS, and more than 200 refereed conferences. Besides that, Prof. Alba has published 11 books, 39 book chapters, and has merited 6 awards to his professional activities. Pr. AlbaвЂ™s H index is 37, with more than 6500 cites to his work.
Algorithm and Proef Vormgeving with HeuristicLab – An Open Source Optimization Environment for Research and Education
Stefan Wagner, University of Applied Sciences Upper Austria, Austria
Gabriel Kronberger, University of Applied Sciences Upper Austria, Austria
HeuristicLab is an open source system for heuristic optimization that features many metaheuristic optimization algorithms (e.g., genetic algorithms, genetic programming, evolution strategies, taboo search, simulated annealing) spil well spil many optimization problems (e.g., traveling salesman, regression, classification, voertuig routing, knapsack, job shop scheduling, simulation-based optimization). It is based on C# and the Microsoft .Netwerk Framework and is used spil development verhoging for several research and industry projects spil well spil for training metaheuristics ter university courses. This tutorial demonstrates how to apply HeuristicLab ter research and education for creating, executing and analyzing metaheuristic optimization algorithms. It includes many interactive live demonstrations ter which it will be shown how to parameterize and execute evolutionary algorithms to solve combinatorial optimization problems spil well spil gegevens analysis problems. The participants will see how to assemble different algorithms and parameter settings to large scale optimization experiments with HeuristicLab’s graphical user interface and how to execute such experiments on multi-core or cluster systems. Furthermore, the proef results will be compared using HeuristicLabвЂ™s interactive charts for visual and statistical analysis. To finish the tutorial, it will be sketched shortly how HeuristicLab can be extended with further optimization problems and how custom-made optimization algorithms can be modeled using the graphical algorithm designer.
Stefan Wagner received his PhD te technical sciences te 2009 from the Johannes Kepler University Linz, Austria. From 2005 to 2009 he worked spil an associate professor for software project engineering and since 2009 spil a total professor for complicated software systems at the University of Applied Sciences Upper Austria, Campus Hagenberg. He is one of the founders of the research group Heuristic and Evolutionary Algorithms Laboratory (HEAL) and is the project manager and head developer of the HeuristicLab optimization environment.
Gabriel Kronberger received his PhD te engineering sciences te 2010 from JKU Linz, Austria, his research interests include genetic programming, machine learning, and gegevens mining and skill discovery. Since 2011 he is professor for gegevens engineering and business intelligence at the University of Applied Sciences Upper Austria, Campus Hagenberg.
PPSN logo by Bogdan FilipiДЌ and Asparuh Mihailov