Genetic Algorithms Digest Tuesday, May 15, 2001 Volume 15 : Issue 18 - Send submissions (articles) to GA-List@gmu.edu. DO NOT send submissions to the listproc@gmu.edu address. - To subscribe send email to listproc@gmu.edu containing the following text in the body of the message: subscribe ga-list - To unsubscribe send email to listproc@gmu.edu containing the following text in the body of the message: unsubscribe ga-list - To change your email address, simply unsubscribe the old address and subscribe the new one. - Send other administrative requests to GA-List@gmu.edu. - You can access back issues, GA code, conference announcements, etc., either through the WWW at http://www.aic.nrl.navy.mil/galist/ or through anonymous ftp at ftp.aic.nrl.navy.mil in /pub/galist. -------------------------------- Today's Topics: - New Book Announcement - NFL and algorithmic complexity - No Free Lunch or Bad Case of Indigestion? - EXTENDED DEADLINE - EA01 - Le Creusot, France, October 29-31, 2001 - IEEE Data Mining 2001: Call for Papers - CFP: IEEE Transactions on IE Special Issue on Soft Computing... -------------------------------- CALENDAR OF GA-RELATED ACTIVITIES: (with GA-List issue reference) CEC2001 Congress on EC, Seoul, Korea May 27-30, 01 (v14n15) ICCS2001 Int Conf on Computational Sci, San Francisco May 28-30, 01 (v14n19) Agents2001 5th Int Conf Autonomous Agents, Montreal May 28-Jun 1, 01 (v14n14) CSCS13 Int. Conf. on Control Sys. and CS, Bucharest May 31-June 3, 01 (v15n4) IC-AI2001 Int Conf on AI, Las Vegas, NV Jun 25-28, 01 (v14n16) SOCO Soft Computing & Intell Sys for Industry, Scotland Jun 26-29, 01 (v14n18) CEF'2001 Sessions on EC in Econ. and Fin., New Haven, CN Jun 28-30, 01 (v15n4) ICML2001 18th Int Conf on Machine Learning, MA Jun 28-Jul 1, 01 (v14n16) AIME01 8th Euro Conf on AI in Medicine, Portugal Jul 1-4, 01 (v14n16) CIMCA2001 Int Conf on Comp Intelligence, Las Vegas Jul 4-6, 01 (v14n19) WOMAII Workshop on Memetic Algorithms, SF, CA Jul 7, 01 (v15n4) IWLCS-2001 4rth Int. Conf. on Classifier Sys., SF, CA Jul 7-8, 01 (v15n10 GECCO2001 Gen & Evolutionary Computation Conf, SF, CA Jul 7-11, 01 (v14n16) TARK VIII 8th Conf Theor Aspects of Ratnlty & Knowl, It Jul 8-10, 01 (v14n16) CIMCA2001 Int. Conf. on Comp. Int.,..., Las Vegas, NV Jul 9-11, 01 (v15n4) IAWTIC2001 Int. Conf. on Int. Agents,..., Las Vegas, NV Jul 9-11, 01 (v15n4) WATT2001 Wrkshp of...Automated Timetabling, Rotterdam Jul 9-11, 01 (v15n8) NASAEH 3rd Wrkshp on Evolvable Hardware, Pasadena, CA Jul 12-14, 01 (v15n2) SCI2001 Evolvable Sys. and Gen. Prog., Orlando, FL USA Jul 22-25, 01 (v15n8) IJCAI-01 WS on Empirical MEthods in AI, Seattle, USA Aug 4, 01 (v15n5) IDAMAP2001 Intelligent Data Analysis in Medicine & Phar Sep 4, 01 (v15n3) FUZZY DAYS Int Conf on Comp Intell, Dortmund, Germany Oct 1-3, 01 (v14n17) ICES2001 4th Int Conf on Evolvable Systems, Tokyo Oct 3-5, 01 (v14n19) MCCS 2001 6th Int. Conf on Msr & Ctrl in Compl. Sys, Ukr Oct. 8-12, 01 (v15n10) IAT2001 2nd Asia Pac Conf on Intell Agent Tech, Japan Oct 23-26, 01 (v14n14) EA01 ÉVOLUTION ARTIFICIELLE 2001, Le Creusot, France Oct 29-31, 01 (v15n5) ICDM01 IEEE Int Conf on Data Mining, Silicon Valley, Nov 29-Dec 2, 01 (v14n14) ANNIE 2001 Smart Eng. Systems Design Conf, StL, MO, USA Nov 4-7, 01 (v15n5) FUZZ-IEEE01 10th IEEE Int Conf on Fuzzy Systems, Austr Dec 2- 5, 01 (v14n20) AMT01 6th Int Conf Active Media Tech, Hong Kong, China Dec 18-20, 01 (v15n16) NF2002 1st Int ICSC Congress on Neuro-Fuzzy, Cuba Jan 15-18, 02 (v14n18) PATAT 2002 4rth Int. Conf. ... Auto. Timetbl., Belgium Aug 21-23, 02 (v15n10) Send announcements of other activities to GA-List@gmu.edu -------------------------------- Sender: Kalyanmoy Deb Subject: New Book Announcement Title: Multi-Objective Optimization Using Evolutionary Algorithms Author: KALYANMOY DEB Publisher: Chichester, UK: Wiley Pages: 496 (Hard Cover) ISBN: 0471 87339 X http://catalog.wiley.com/remtitle.cgi?isbn=047187339X&country=826 The above book is now available from John Wiley & Sons (http://www.wiley.co.uk). One of the niches of evolutionary algorithms in solving search and optimization problems is the elegance and efficiency in which they can solve multi-objective optimization problems. Multi-objective optimization deals with multiple and often conflicting objectives, thereby resulting in a set of optimal solutions instead of a single optimal solution. This book is the first comprehensive book introducing multi-objective optimization, classical multi-objective optimization methods, evolutionary algorithms, and immediate research topics in the emerging field of multi-objective evolutionary algorithms. The highlights of the book: o Comprehensive coverage of the growing area of multi-objective evolutionary algorithms o Carefully introduces each algorithm with examples and in-depth discussion o Includes a number of real-world problems from engineering and sciences o Includes discussion of advanced topics and future research o Includes an extensive reference list of current research studies o Accessible to those with limited knowledge of classical multi-objective optimization and evolutionary algorithms The integrated presentation of theory, algorithms and examples will benefit those working and researching in the areas of optimization, optimal design and evolutionary computing. This text provides an excellent introduction to the use of evolutionary algorithms in multi-objective optimization, allowing use as a graduate course text or for self-study. Multi-Objective Optimization Using Evolutionary Algorithms by Kalyanmoy Deb CONTENTS: Foreword by David E. Goldberg Preface Prologue Multi-Objective Optimization Classical Methods Evolutionary Algorithms Non-elitist multi-objective evolutionary algorithms Elitist multi-objective evolutionary algorithms Constrained multi-objective evolutionary algorithms Salient Issues of multi-objective evolutionary algorithms Applications of multi-objective evolutionary algorithms Epilogue References Index -------------------------------- Sender: Thomas English Subject: NFL and algorithmic complexity Considerations of algorithmic complexity in "no free lunch" (NFL) depend critically upon whether one is treating optimization or learning [1]. Here I limit my comments to optimization. It has been suggested that bounding the algorithmic complexity of test functions results in a "free nibble"--i.e., a small superiority for some optimizers over others. In reality, the bound does not suffice. NFL is a property of the distribution of functions, and can be exhibited even if the distribution is defined on a set of functions of very low algorithmic complexity. For instance, there is no free lunch if functions are drawn uniformly from the set of all "needle-in-a-haystack" functions mapping exactly one domain element to 1 and all other domain elements to 0 [2]. The functions in this set are of very low algorithmic complexity, so no complexity bound of practical interest rules out the possibility of an NFL distribution of test functions. It is interesting to note that the uniform distribution on needle-in-a-haystack functions is both the hardest to maximize and the easiest (of all distributions excluding constant functions) to minimize. That is, a single NFL distribution can be exceedingly benign or perverse, depending upon the optimization objective. Thomas English Tom.English@ieee.org [1] English, T. M. 2000. "Optimization is easy and learning is hard in the typical function," Proceedings of 2000 Congress on Evolutionary Computation: CEC00, pp. 924-931 (http://members.door.net/tmenglish/cec2000.pdf). [2] English, T. M. 1996. "Evaluation of evolutionary and genetic optimizers: No free lunch," Evolutionary Programming V: Proc. of the Fifth Ann. Conf. on Evolutionary Programming, L. J. Fogel, P. J. Angeline, and T. Baeck, eds., pp 163-169. -------------------------------- Sender: "John Koza" Subject: No Free Lunch or Bad Case of Indigestion? Hello All: The recent postings to the GA-List by Ralph Hartley and Thomas Jansen on the "No Free Lunch" theorem (or the "No Lunch" theorem as Hartley would call it) reminded me of an experiment done in the early 1990's and reported in the first Genetic Programming book (1992, MIT Press). This experiment involved one of the few areas where the universe of "all possible problems" can be readily identified and realistically tackled by computer experimentation, namely Boolean functions. At the time, I recall discussing with various people whether there was some general "conservation" principle operating over the universe of "all possible problems"; however, none of us followed up on this vague conjecture at the time since this idea was entirely peripheral to the work that we were doing. The extremely clever actual later work on NFL (occurring in the mid-1990's) provided insightful clarification of what was going on here. Here are some quotes from the 1992 book concerning the 3-argument and 2-argument Boolean functions. Experiments with Boolean functions of more than three arguments could, of course, be easily conducted today. The goal will be to compare the number of individuals that must be processed by genetic programming to the number of individuals that must be processed in a specified blind random search. We will first focus on the functions of three Boolean arguments and one Boolean output, because the total number of such functions is small enough to permit exhaustive examination with the available computational resources. We then repeat the process for the functions of two Boolean arguments. We will reach the following conclusions for both classes of functions: o Genetic programming can produce a solution for 100% of the functions in the class of functions. o Genetic programming finds a solution after processing fewer individuals than a blind random search, except for the degenerate functions and manifestly simple functions in the class of functions under consideration. For these degenerate functions and these simple functions, genetic programming finds a solution after processing the same number of individuals or slightly more individuals (owing to its overhead) than a blind random search. o The advantage of genetic programming over blind random search generally increases as the functions become more complex. In other words, genetic programming does better on the harder functions of each class. ... For rule 247 and all 28 rules between rows 3 and 30 of table 9.4, the difference in logarithms (column 8) is slightly negative and the point plotted in figure 9.2 appears slightly above the 45° line. That is, genetic programming finds a solution individual by processing more individuals than the blind random search for those rules. Many of these rules, such as (AND D2 D0), are degenerate in that they do not involve all three input arguments; the others, such as (AND D2 (AND D1 D0)), are comparatively simple Boolean rules. These 28 rules are apparently too simple for genetic programming to handle efficiently (because of the overhead associated with genetic programming). ... In summary, for the Boolean functions with three arguments, as many or slightly more individuals must be processed by genetic programming than blind random search in order to find the degenerate and very simple functions, but considerably fewer individuals must be processed by genetic programming for the majority of the functions, including the harder functions (and, notably, the odd and even 3-parity functions). Moreover, the advantage of genetic programming generally increases for the harder functions. The point that Hartley very correctly makes in his posting is that the "problems" for which the various search techniques do worse than random are, as a practical matter, the ones which no one is concerned about. John R. Koza Consulting Professor Biomedical Informatics Department of Medicine Medical School Office Building (MC 5479) Stanford University Stanford, California 94305-5479 Consulting Professor Department of Electrical Engineering School of Engineering Stanford University Phone: 650-941-0336 Fax: 650-941-9430 E-Mail: koza@stanford.edu http://www.smi.stanford.edu/people/koza http://www.genetic-programming.org http://www.genetic-programming.com -------------------------------- Sender: Evelyne.Lutton@inria.fr Subject: EXTENDED DEADLINE - EA01 - Le Creusot, France, October 29-31, 2001 ******************************************************************** EXTENDED DEADLINE FOR PAPERS SUBMISSION TO MAY 25 ******************************************************************** CALL FOR PAPERS ÉVOLUTION ARTIFICIELLE 01 - EA'01 October 29-31, 2001 Université de Bourgogne, Le Creusot, France EA'01 is the 5th Conference on Evolutionary Computation organized in France. Following EA'94 in Toulouse, EA'95 in Brest, EA'97 in Nîmes, and EA'99 in Dunkerque, the Conference will be held in Le Creusot, hosted by the Université de Bourgogne. As usual, the conference will benefit from its specificity: keeping away from Paris will allow for both very low registration fees (less than 150 EUROS) and the discovery of another French region, renowned for its gastronomy and its very famous wines: Burgundy. Authors are invited to present original work relevant to simulated evolution, including but not limited to: Evolutionary Optimization and Learning - Theory of Evolutionary Computation - Artificial Life - Population Dynamics - Implementation and Real-World Application of evolutionary paradigms. PAPERS All submitted papers will be reviewed by three members of the Program Committee. All accepted papers will be presented at the Conference (single track oral presentation of 20mn with 10mn for questions, and discussion, in order to favor interactions among participants), and published in the conference proceedings, available at the conference. If quality permits, a selection of the best papers will further be published by an international publisher, as happened for the previous issues (Springer Verlag Lecture Notes in Computer Science 1063, 1363 and 1829). Electronic submission of the papers (12 pages maximum in Springer lncs format, i.e. 11pt Times Roman font, 17x25cm text), is strongly encouraged. The use of LaTeX is strongly recommended. A model style file can be obtained on the conference Web site, where Gill Bates' fans can also find a Word template. LANGUAGE The official language of the Conference is French, but presentations in English are strongly encouraged --French researchers whose English is not yet fluent will find a unique friendly environment to practice their scientific English. Papers can be submitted in French or English. However, to insure a large international visibility of the presented work, the final revision of the papers accepted for publication in the international volume will need to be written in English. CONTACT Chantal LABEILLE (labeille@cheverny.u-bourgogne.fr) Conférence EA'01 Centre Universitaire Condorcet 720 Ave de l'Europe - 71200 Le Creusot - France Tel: +33 (0)1 03.85.77.00.77 - Fax: +33 (0)1 03.85.55.84.58 e-mail: ea01@cmapx.polytechnique.fr - http://www.cmap.polytechnique.fr/ea01 CALENDAR EXTENDED Deadline for paper submission May 25 2001 Notifications of acceptance July 1 2001 Final version for conference handouts September 15 2001 Conference October 29-31 2001 [ ... Modified for brevity by moderator ... ] -------------------------------- Sender: Ning Zhong Subject: IEEE Data Mining 2001: Call for Papers ---------------------------------------------------------------------- ICDM '01: The 2001 IEEE International Conference on Data Mining Sponsored by the IEEE Computer Society ---------------------------------------------------------------------- Doubletree Hotel, San Jose, California, USA November 29 - December 2, 2001 Home Page: http://kais.mines.edu/~xwu/icdm/icdm-01.html INVITED SPEAKERS: Jerome H. Friedman, Stanford University, USA Jim Gray (The 1999 Turing Award Winner), Microsoft Research, USA Pat Langley, Daimler-Benz Research & Technology Center, USA Benjamin W. Wah (IEEE Computer Society President), UIUC, USA CORPORATE SPONSORS: Blue Martini Software, San Mateo, California; Insightful Corporation, Seattle, Washington; NARAX Inc., Golden, Colorado; Springer-Verlag, New York, New York; StatSoft Inc., Tulsa, Oklahoma Call for Papers *************** The 2001 IEEE International Conference on Data Mining (ICDM '01) provides a forum for the sharing of original research results and practical development experiences among researchers and application developers from different data mining related areas such as machine learning, automated scientific discovery, statistics, pattern recognition, knowledge acquisition, soft computing, databases and data warehousing, data visualization, and knowledge-based systems. The conference seeks solutions to challenging problems facing the development of data mining systems, and shapes future directions of research by promoting high quality, novel and daring research findings. As an important part of the conference, the workshops program will focus on new research challenges and initiatives. Topics of Interest ================== Topics related to the design, analysis and implementation of data mining theory, systems and applications are of interest. These include, but are not limited to the following areas: - Foundations and principles of data mining - Data mining algorithms and methods in traditional areas (such as classification, clustering, probabilistic modeling, and association analysis), and in new areas - Data and knowledge representation for data mining - Modeling of structured, textual, temporal, spatial, multimedia and Web data to support data mining - Complexity, efficiency, and scalability issues in data mining - Data pre-processing, data reduction, feature selection and feature transformation - Statistics and probability in large-scale data mining - Soft computing (including neural networks, fuzzy logic, evolutionary computation, and rough sets) and uncertainty management for data mining - Integration of data warehousing, OLAP and data mining - Man-machine interaction in data mining and visual data mining - Artificial intelligence contributions to data mining - High performance and distributed data mining - Machine learning, pattern recognition and automated scientific discovery - Quality assessment and interestingness metrics of data mining results - Process centric data mining and models of data mining process - Security and social impact of data mining - Emerging data mining applications, such as electronic commerce, Web mining and intelligent learning database systems Conference Publications and ICDM Best Paper Awards ================================================== High quality papers in all data mining areas are solicited. Papers exploring new directions will receive a careful and supportive review. All submitted papers should be limited to a maximum of 6,000 words (approximately 20 A4 pages), and will be reviewed on the basis of technical quality, relevance to data mining, originality, significance, and clarity. Accepted papers will be published in the conference proceedings by the IEEE Computer Society Press. A selected number of ICDM '01 accepted papers will be expanded and revised for possible inclusion in the Knowledge and Information Systems journal (http://kais.mines.edu/~kais/) by Springer-Verlag. ICDM Best Paper Awards will be conferred on the authors of the best papers at the conference. Important Dates =============== June 15, 2001 Paper submissions. July 31, 2001 Acceptance notices. August 31, 2001 Final camera-readies. Nov 29 - Dec 2, 2001 Conference. All paper submissions will be handled electronically. Detailed instructions are provided on the conference home page at http://kais.mines.edu/~xwu/icdm/icdm-01.html. Conference Chair: ================= Xindong Wu, Colorado School of Mines, USA (xindong@computer.org) [ ... Modified for brevity by moderator ... ] Further Information =================== Dr. Xindong Wu Dept. of Mathematical and Computer Sciences, Colorado School of Mines, 1500 Illinois Street, Golden, Colorado 80401, USA. Telephone: +1-303-273-3874 Facsimile: +1-303-273-3875 E-mail: xindong@computer.org -------------------------------- Sender: Yaochu_Jin@de.hrdeu.com Subject: CFP: IEEE Transactions on IE Special Issue on Soft Computing Techniques in IVS Call for Papers IEEE Transactions on Industrial Electronics Special Issue on Soft Computing Techniques in Intelligent Vehicle Systems Scope The information age we are embracing is imposing great challenges to the Intelligent Vehicle Systems (IVS). Modern car drivers expect to be able to drive safely while exchanging information with the outside world. Vehicle safety technologies, such as collision warning, driver assistance and autonomous driving, as well as injury reduction in case of an accident are the basic concerns of intelligent vehicle systems. Information and connectivity is another essential aspect. Intelligent vehicle systems are supposed to be able to provide filtered information about local traffic conditions, navigation, and weather conditions and provide useful suggestions. With the help of Internet and telecommunication technologies, drivers can check emails, browsing Internet and even handle business without stepping out of the car. To meet the increasing demand for safety and connectivity, intelligent vehicle systems need to have stronger capability of understanding the environment, learning from the history, and making correct decisions with uncertain, partial or imprecise information. Soft Computing is an emerging field that consisting of complementary elements of Fuzzy Logic, Neural Computing, Evolutionary Computation, Machine Learning and Probabilistic Reasoning. Due to their strong learning and cognitive ability and good tolerance of uncertainty and imprecision, Soft Computing techniques have found wide applications in Intelligent Vehicle Systems. This Special Issue will be dedicated to the publication of the latest advancements in theory and application of Soft Computing techniques to intelligent vehicle systems. Topics may include but are not limited to: Adaptive Cruise Control Collision Avoidance and Obstacle Detection Driver Behavior Modeling and Monitoring Lane Detection and Tracking Optimization of Vehicle Safety Equipments In-Vehicle Navigation and Communication Human-Machine Systems for IVS Driver Information Systems Traffic Modeling and Control Submission Four hard copies of each submitted papers should be sent to the one of the Guest Editors for the author's region at the addresses below. Electronic submissions in postscript or pdf format are encouraged. Submitted papers should be in the IEEE TIE format with a title page including a complete mailing address for each author plus an abstract of the paper. Please also email a copy of the title page in plain text to one of the Guest Editors. More information on style guidelines for a submission can be obtained at the IEEE Transactions on Industrial Electronics' Website: http://www.trans-ie.uni-wuppertal.de/ Submission deadline 31 July 2001 Notification of accepta nce 30 September 2001 Guest Editors Sam Kwong Department of Computer Science City University of Hong Kong 83 Tatchee Avenue Kowloon, Hong Kong China Email: cssamk@cityu.edu.hk Yaochu Jin Future Technology Research Honda R&D Europe (D) GmbH Carl-Legien-Strasse 30 63073 Offenbach/Main Germany Email: yaochu_jin@de.hrdeu.com -------------------------------- End of Genetic Algorithms Digest ********************************