Dece

mb

er 0

6-0

8, 2

00

4K

itakyu

shu

, Jap

an

Advanced Programme

HIS’04 Sponsors

In cooperation with

IEEE

Technical Sponsors

BMFSA

HIS'04 is supported by the City of Kitakyushu

HIS'04 Conference Organization

Honorary ChairShunichi Amari, RIKEN Brain Science Institute, Japan General ChairsMario Köppen, Fraunhofer IPK FHG, GermanyDavid Corne, University of Exeter, UKAjith Abraham, Chung-Ang University, Korea Program ChairsShuji Hashimoto, Waseda University, JapanMasumi Ishikawa, Kyushu Institute of Technology, Japan International Co-ChairsYasuhiko Dote, Muroran Institute of Technology, JapanLakhmi Jain, University of South Australia, AustraliaKate Smith, Monash University, Australia Special Event ChairKatrin Franke, Fraunhofer IPK FHG, Germany Publicity ChairVitorino Ramos, CVRM - IST, Portugal

Akira Asano, Hiroshima University, JapanAladdin Ayesh, De Montfort University, UKAndre Leon, University of Sao Paulo Sao Carlos, BrazilAndreas Koenig, University of Kaiserslautern, GermanyAndrew Sung, New Mexico Tech, USAAntonio Braga, UFMG, BrazilAureli Soria Frisch, Fraunhofer IPK Berlin, GermanyBernard de Baets, Ghent University, BelgiumBernard Grabot, LGP/ENIT, FranceBruno Apolloni, Universita degli Studi di Milano, ItalyCarlos Coello, CINVESTAV-IPN, MexicoChristian Veenhuis, Fraunhofer IPK Berlin, GermanyCosta-Branco P J, Instituto Superior Tecnico Lisboa, PortugalDamminda Alahakoon, University of Colombo, AustraliaEiji Uchino, University of Yamaguchi, JapanElisa Bertino, Purdue University, USAEmilia Barakova, Brain Science Institute RIKEN, JapanErkki Oja, Helsinki University of Technology, FinlandEtienne Kerre, Ghent University, BelgiumFabio Abbattista, Universita di Bari, ItalyFrancisco Herrera, University of Granada, SpainFrancisco Pereira, Universidade de Coimbra, PortugalFrank Hoffmann, University of Dortmund, GermanyFrank Klawonn, TFH Braunschweig/Wolfenbuettel, GermanyGail Carpenter, Boston University, USAGary Fogel, Natural Selection Inc., USAGancho Vachkov, Kagawa University, JapanGheorghe Tecuci, George Mason University, USAGiovanni Semeraro, Universita di Bari, ItalyGuenther Raidl, Vienna University of Technology, AustriaHalina Kwasnicka, Wroclaw University, PolandHisao Ishibuchi, Osaka Prefecture University, JapanJanina Jakubczyc, Wroclaw University of Economics, PolandJanos Abonyi, University of Veszprem, HungaryJarno Tanskanen, University of Kuopio, FinlandJavier Ruiz-del-Solar, Universidad de Chile, ChileJerzy Grzymala-Busse, University of Kansas, USAJiming Liu, Hong Kong Baptist University, Hong KongJose Mira, UNED, SpainKaori Yoshida, Kyushu Institute of Technology, Japan

Additional Reviewers: Bruno Feres de Souza, Debora Maria Rossi de Medeiros, Simone Bassis, Fabio Fumagalli, Andrea Brega, Anna Morpurgo, Balazs Feil, Janos Madar, Renato Krohling, Noriaki Suetake,Steven Schockaert, Filippo Lanubile, Nicola Fanizzi, Luigi Iannone, Ignazio Palmisano, Corrado Mencar, Ciro Castiello Denitsa Apostolova, William Wagner, Marcel Barbulescu, Ping Shyr, Vu Le, DorinMarcu, Jiaqi Wang, Li Lin, Zhenxin Qin, Fengjie Wu

Katrin Franke, Fraunhofer IPK Berlin, GermanyKazuyuki Murase, Fukui University, JapanLipo Wang, Nanyang Technological University, SingaporeLongbing Cao, University of Technology, Sydney, AustraliaLouis Vuurpijl, NICI Nijmegen, The NetherlandsLuis Magdalena, Universidad Politecnica de Madrid, SpainMarcin Paprzycki, Oklahoma State University, USAMarley Vellasco, PUC-Rio, BrasiliaMartin Stytz, Airforce Research Laboratory, USAMatjaz Gams, Jozef Stefan Institute Ljubljana, SloveniaMika Sato-Ilic, University of Tsukuba, JapanMiroslav Karny, Academy of Sciences, Czech RepublicNadia Nedjah, State University of Rio de Janeiro, BrazilOlgierd Unold, Wroclaw University of Technology, PolandRajkumar Roy, Cranfield University, UKRichard Weber, Universidad de Chile, ChileRobert Howlett, University of Brighton, UKRonald R. Yager, Iona College, USASaman Halgamuge, The University of Melbourne, AustraliaSankar K. Pal, Indian Statistical Institute, IndiaShusaku Tsumoto, Shimane Medical University, JapanStefan Wermter, University of Sunderland, UKSung-Bae Cho, Yonsei University, KoreaSuthikshn Kumar, Infotech, IndiaTim Hendtlass, Swinburne University of Technology, AustraliaTom Gedeon, Australian National University, AustraliaUdo Seiffert, IPK Gatersleben, GermanyUwe Zimmer, Australian National University, AustraliaVasile Palade, Oxford University, UKVioletta Galant, Wroclaw University of Economics, PolandWiliam Browne, University of Reading, UKWilliam Langdon, University College, London, UKWitold Pedrycz, University of Alberta, CanadaXiao-Zhi Gao, Helsinki University of Technology, FinlandYanqing Zhang, Georgia State University, USAYasuhiko Dote, Muroran Institute of Technology, JapanYoshiyasu Takefuji, Keio University, JapanYuzo Hirai, University of Tsukuba, JapanZensho Nakao, University of the Ryukyus, Japan

Local Organizing ChairsEmilia Barakova, RIKEN Brain Science Institute, JapanKaori Yoshida, Kyushu Institute of Technology, JapanHirokazu Yokoi, Kyushu Institute of Technology, Japan International Steering CommitteeDavid Fogel, Natural Selection Inc., USAJanusz Kacprzyk, Polish Academy of Sciences, PolandVasile Palade, University of Oxford, UKTakeshi Yamakawa, Kyushu Institute of Technology, JapanJavier Ruiz-del-Solar, Universidad de Chile, ChileJae Oh, Syracuse University, USAMarley Vellasco, PUC-Rio, BrazilAntony Satyadas, IBM, USA

Publication ChairMarcin Paprzycki, Oklahoma State University, USA

International Program Committee

General Chairs Welcome Message

We are very pleased to once again welcome our colleagues to the Fourth International Conference onHybrid Intelligent Systems (HIS’04) at the Kitakyushu International Conference Center, Kitakyushu,Japan. The HIS series of conferences attracts a wide range of interests with the development of thenext generation of intelligent systems and its various practical applications. A fundamental stimulus tothe investigation of hybrid intelligent systems is the awareness in the academic communities thatcombined approaches will be necessary if the real challenges in artificial intelligence are to be solved.Current research interests in this field focus on integration of the different computing paradigms such asfuzzy logic, neurocomputation, evolutionary computation, probabilistic computing, intelligent agents,machine learning, and other intelligent computing frameworks. The phenomenal growth of hybridintelligent systems and related topics has created the need for this International conference as a venueto present the latest research. HIS’04 builds on the success of last year’s. HIS’03, which was held inMelbourne, Australia, 14-17, December 2003 and attracted participants from over 32 countries. Asevident from the general philosophy of HIS conferences, we have a focus on interdisciplinaryapproaches and the global strategy to bring together research scientists from the various disciplinesrelated to hybrid intelligent systems. HIS’04, the Fourth International Conference on Hybrid IntelligentSystems, in Kitakyushu, Japan, December 05-08, 2004, addresses the following important themes:

* Theoretical advances in hybrid intelligent system architectures

• Interactions between neural networks and fuzzy inference systems

• Hybrid learning techniques (supervised/unsupervised/reinforcement learning)

• Artificial neural network optimization using global optimization techniques

• Fuzzy inference system optimization using global optimization algorithms

• Hybrid systems involving support vector machines, rough sets, Bayesian networks,probabilistic reasoning, minimum message length, etc.

• Hybrid computing using neural networks - fuzzy systems - evolutionary algorithms

• Hybrid optimization techniques (evolutionary algorithms, simulated annealing, tabu searchetc.)

• Hybrid of soft computing and statistical learning techniques

• Integration with Intelligent agents (architectures, environments, adaptation/learning andknowledge management)

• Hybrid models using inductive logic programming, logic synthesis, grammatical inference,case-based reasoning etc.

* Hybrid approaches and applications

• Biomimetic applications

• Bioinformatics

• Robotics and automation

• Web intelligence

• Image and signal processing

• Data mining

• Behavioral simulations

• Affective computing

• Control and automation

• Multi-agent systems

• Knowledge management

• Communication and networking

• Business systems and financial engineering.

• Power engineering

HIS’04 is technically co-sponsored by IEEE Systems Man and Cybernetics Society, International FuzzySystems Association, The World Federation on Soft Computing, Japan SOciety for Fuzzy Theory andintelligent informatics (SOFT), Biomedical Fuzzy Systems Association (BMFSA, Japan) and is organizedin cooperation with the IEEE Computational Intelligence Society. HIS’04 is supported by the City ofKitakyushu, Japan and is co-hosted by the Riken Brain Science Institute, Tokyo, Japan and KyushuInstitute of Technology, Japan.

The HIS’04 program committee represented 22 countries on 5 continents and authors submitted papersfrom 26 countries on 6 continents. This certainly attests to the widespread, international importance ofthe theme of the conference. Each paper was peer reviewed by at least three or more programcommittee members and based on the recommendations of the reviewers, 70 regular papers and 14poster papers were included in the final Programme. We would like to thank the HIS’04 internationalprogram committee and the additional reviewers for providing the reviews in time. Our special thanksalso go to all the plenary speakers for providing the very interesting and informed talks to catalyzesubsequent discussions.

Our special thanks to Ms. Jeniferdawn Cantarella of IEEE Computer Society Press for all the supportand help related to the production of this important scientific work. Finally, we would like to express oursincere gratitude to all the authors and local organizing committees that have contributed towards thesuccess of this conference. We look forward to seeing you in Kitakyushu, Japan during HIS’04,December 05-08, 2004.

November 30, 2004 Mario Köppen, Fraunhofer IPK, Berlin, GermanyDavid Corne, University of Exeter, UKAjith Abraham, Chung-Ang University, Seoul, Korea

HIS’04 General Chairs

Program Chairs Welcome Message

Welcome to Kitakyushu, Japan!

The Fourth International Conference on Hybrid Intelligent Systems, HIS'04, has evolved out of a seriesof successful HIS conferences, the last one being held in Melbourne, Australia. The objectives of thisinternational meeting are to increase the awareness of the research area with a broad spectrum ofhybrid techniques, to bring together AI researchers from around the world to present their cutting-edgeresearch results, to discuss the current issues in HIS research, to get a perspective view on futuredevelopments, to foster international collaboration, and as a result to further advance the state of the artof the field.

HIS’04 solicited papers on 1) Theoretical Advances in Hybrid Intelligent System Architectures and 2)Hybrid Approaches and Applications in various fields. Reflecting a surge of interest in this field, about180 papers were submitted and from these about 80 high-quality papers are selected for presentation bycompetent reviewers. HIS’04 also provides three Organized Technical Sessions proposed bydistinguished researchers in each field. It is our great pleasure to have Dr. Una May O'Reilly, Prof.Witold Pedrycz, Prof. Toshio Fukuda, Prof. Nikola Kasabov and Prof. Sung-Bae Cho as plenaryspeakers at the Conference.

We would like to express our sincere gratitude to all the authors for submission, and to the members ofthe program committee for their careful and prompt reviews. We would also like to thank all themembers of the International and Local Committees and co-sponsoring societies. All their efforts wereindispensable to make HIS’04 possible. We would also express our gratitude to City of Kitakyushu forthe valuable supports.

Finally, we would like to thank all the participants of HIS’04. Let’s get together to talk about the cutting-edge topics of HIS and enjoy!

November 29, 2004 Shuji Hashimoto, Waseda University, JapanMasumi Ishikawa, Kyushu Institute of Technology, Japan

HIS’04 Program Chairs

HIS’04 - Programme at a Glance

The program for this conference consists of a morning session, and an afternoon session, each of 3-4 hoursduration arranged in 2 parallel tracks over three days (Monday, December 06, Tuesday, December 07 andWednesday, December 08), 5- 5 papers of 20 minutes duration each will be presented in each session. There willbe 70 papers and 14 posters presented at the conference. Five plenary lectures are presented by distinguishedresearchers in the field of hybrid intelligent systems. Conference banquet will be held on Tuesday after the eveningsession at the conference center.

Venue of HIS'04 will be the Kitakyushu International Conference Center.

Venue Address:Kitakyushu International Conference Center3-9-30, Asano, Kokurakita-ku, Kitakyushu 802, JAPANTel (093) 541-5931/ Fax (093) 541-5928

Overview of HIS’04 Conference Programme

Monday, December 06, 2004

08:30-16:00 Registration

09:00-09:30 Welcome greetings

09:30-10:30 Plenary Session I (Witold Pedrycz, University of Alberta, Canada)

10:30-11:00 Tea/Coffee Break

11:00–13:00 Session 1.1. Machine learning and heuristics I Session 1.2. Web intelligence and data mining

13:00–14:00 Lunch

14:00-15:00 Plenary Session II (Sung-Bae Cho, Yonsei University, Korea)

15:00-15:30 Tea/Coffee Break

15:30-17:00 Session 1.3. Soft Computing Systems I Session 1.4. Agent systems, design & architectures I

17:00-18:00 Poster Session

Tuesday, December 07, 2004

08:30-16:00 Registration

09:30-10:30 Plenary Session III (Nikola Kasabov, KEDRI, Auckland, New Zealand)

10:30-11:00 Tea/Coffee Break

11:00–13:00 Session 2.1. Agent systems, design & architectures II Session 2.2. Knowledge Processing & Management I

13:00–14:00 Lunch

14:00-15:00 Plenary Session IV (Una May O’Reilly, MIT, USA)

15:00–15:30 Tea/Coffee break

15:30-16:30 Panel Discussion: Future of Hybrid Intelligent Systems

16:30-18:30 Session 2.3. Anticipatory processing Session 2.4. Machine learning and heuristics II

19:00-21:00 Conference Banquette

Wednesday, December 08, 2004

09:30-10:30 Plenary Session V (Toshio Fukuda, Nagoya University, Japan)

10:30-11:00 Tea/Coffee Break

11:00-13:00 Session 3.1. Knowledge Proc. & Management II Session 3.2. Soft Computing Systems II

13:00: 14:00 Lunch

14:00-15:30 Session 3.3. Features, classification & clustering Session 3.4. Image and Signal Processing

15:30-16:00 Tea/Coffee break

16:00-18:00 Session 3.5. Knowledge Proc. & Management III Session 3.6. Optimization

18:00: 18:15 Concluding remarks

HIS’04 Programme

Monday, December 06, 20048:30 AM – 4:00 PM

Registration

Monday, December 06, 9:00 AM – 9:30 AM Welcome Greetings

Monday, December 06, 9:30 AM – 10:30 AM (Plenary Session - I)

Granular Modeling: The Synergy of Granular Computing and Fuzzy LogicWitold Pedrycz, University of Alberta, Canada

Tea/Coffee Break: 10:30 – 11:00 AM

Parallel Technical Sessions: Monday, December 06, 11:00 AM – 01:00 PM

Session 1.1 (Room A): Machine Learning and Heuristics IChair: Bernardete RibeiroCo-Chair: Joarder Kamruzzaman

Margin-based Active Learning using Background Knowledge in Text MiningCatarina Silva, Bernardete Ribeiro

Hybridising Rule Induction and Multi-Objective Evolutionary Search for Optimising Water Distribution SystemsLaetitia Jourdan, David Corne, Dragan Savic, Godfrey Walters

Robotic Hand-Eye Coordination: From Observation to ManipulationShahram Jafari, Ray Jarvis

An Intelligent Model for Reconstruction of Stance Time from Faulty Gait RecordingsJoarder Kamruzzaman, Rezaul Begg

Comparisons between Heuristics Based on Correlativity and Efficiency for Landmarker GenerationDaren Ler, Irena Koprinska, Sanjay Chawla

Using Association Rules for Completing Missing DataChih-Hung Wu and Chian-Huei Wun

Session 1.2 (Room B): Web Intelligence and Data MiningChair: Nik Kasabov

Image Clustering System on WWW using Web TextsWataru Sunayama, Akiko Nagata, and Masahiko Yachida

Application of Machine Learning Techniques to Web-Based Intelligent Learning Diagnosis SystemChenn-Jung Huang, Ming-Chou Liu, San-Shine Chu, Chin-Lun Cheng

Application of Support Vector Machines to Admission Control for Proportional Differentiated Services EnabledInternet Servers

Chenn-Jung Huang, Chih-Lun Cheng

Intelligent Agent to Support Design in Supply Chain Based on Semantic Web ServiceIncheon Paik, Shinjiro Takami, Yuu Watanabe

Improvement of Non-Linear Mapping Computation for Dimensionality Reduction in Data Visualization andClassification

Kuncup Iswandy and Andreas König (Senthil K. Laksmanan)

Hybrid learning scheme for Data Mining ApplicationsT.Ravindra Babu, M.Narasimha Murty, V.K.Agrawal

Lunch Break: 1:00 PM – 02:00 PM

Monday, December 06, 2:00 PM – 3:00 PM (Plenary Session - II)

Hybrid Intelligent Techniques for Intelligent Personal Assistant in Digital ConvergenceSung-Bae Cho, Yonsei University, Seoul, Korea

Tea/Coffee Break: 3:00 – 03:30 PM

Parallel Technical Sessions: Monday, December 06, 03:30 PM – 05:00 PM

Session 1.3 (Room A): Soft Computing Systems IChair: Shuji Hashimoto

Improving a Pittsburgh Learnt Fuzzy Rule Base Using Feature Subset SelectionPablo A. D. Castro, Daniel M. Santoro, Heloisa A. Camargo, Maria C. Nicoletti

Ensemble of Linear Perceptrons with Confidence Level OutputPitoyo Hartono, Shuji Hashimoto

First-Order Logical Neural NetworksThanupol Lerdlamnaochai and Boonserm Kijsirikul

Performance Studies on KBANNR.P. Jagadeesh Chandra Bose, G. Nagaraja

Artificial Neural Networks Applied to Long-term Electricity Demand ForecastingMostafa Al Mamun, Ken Nagasaka

Session 1.4 (Room B): Agent Systems, Design and Architectures IChair: Jae C. OhCo-Chair: Wenyu Qu

Mobile Agent-Based Execution ModellingWenyu Qu and Hong Shen

Agent-based Modelling: A Case Study in HIV EpidemicEyob Teweldemedhin, Tshilidzi Marwala, Conrad Mueller

Cooperative Game Theory within Multi-Agent Systems for Systems SchedulingDerek Messie, Jae C. Oh

A Rationality-based Modeling for Coalition SupportJae C. Oh, Nathaniel Gemelli and Robert Wright

Poster Session: 05:00 – 06:00 PM

Poster Presentations

An Effective Machine Learning Algorithm using Momentum SchedulingEunmi Kim, Baeho Lee

Design of Robust PID Controller With Disturbance Rejection For Motor Using Immune AlgorithmDong Hwa Kim, Jae Hoon Cho

Robust Motor Control Using clonal selection of Immune Algorithm Based MultiobjectiveDong Hwa Kim, Jae Hoon Cho, Hwan Lee

Realizing Small Involuntary Movements in Reading EnglishYukio Ishihara and Satoru Morita

A Short Study on Dynamical Properties of Population based Incremental Incremental LearningR. Rastegar, M. R. Meybodi

Fuzzy Reasoning in Sparse Fuzzy Rule BasesBaowen Wang, Xia Li, Wenyuan Liu and Yan Shi

Stigmergy in Multi Agent Reinforcement LearningRaghav Aras, Alain Dutech, François Charpillet

Field-Programmable Analog Filters Array with Applications for Fuzzy Inference SystemsSilviu Ionita, Emil Sofron

A Waterfall Model for Knowledge Management and Experience ManagementZhaohao Sun

An Attitude Based Cooperative Negotiation ModelMadhu Goyal

Anatomy of Swarms for Clustering DataAlfred Ultsch

Investigation on the Threshold Setting of Intracellular Calcium Ion Concentration in Hippocampal CA3 NetworkModel for Synaptic Plasticity

Kazuumi Tashiro, Noriaki Suetake and Eiji Uchino

Reinforcement Learning of Player Agents in RoboCup Soccer SimulationAbhinav Sarje, Amit Chawre, Shivashankar B. Nair

Segmentation of High Resolution Satellite Images by Direction and Morphological FiltersTomoko Tateyama, Xian Yan Zeng, Zensho Nakao, Yen-Wei Chen

Tuesday, December 07, 20048:30 AM – 4:00 PM

Registration

Tuesday, December 07, 9:30 AM – 10:30 AM (Plenary Session - III) Discovering Rules of Adaptation and Interaction: From Molecules and Gene Interaction to Brain FunctionsNikola Kasabov, KEDRI, Auckland, New Zealand

Tea/Coffee Break: 10:30 – 11:00 AM

Parallel Technical Sessions: Tuesday, December 07, 11:00 AM – 01:00 PM

Session 2.1 (Room A): Agent Systems, Design and Architectures IIChair: Una May O'ReillyCo-Chair: Asha Rao

Integrative Early Requirements Analysis for Agent-Based SystemsLongbing Cao, Chengqi Zhang, Dan Luo, Wanli Chen, Neda Zamani

Motivation Based Behavior in Hybrid Intelligent Agents for Intention Reconsideration Process in Vessel BerthingApplications

Prasanna Lokuge and Damminda Alahakoon

Reinforcement Learning_Hierarchical Neuro-Fuzzy Politree Model for Control of Autonomous AgentsKarla Figueiredo, Marley Vellasco, Marco Pacheco, Flávio Souza

An Image-Guided Heuristic for Planning an Exhaustive EnumerationJoselito J. Chua and Asha Rao

A Novel Adaptive Decision Making Agent Architecture Inspired by Human Behavior and Brain Study ModelsL. K. Wickramasinghe, L.D. Alahakoon

Session 2.2 (Room B): Knowledge Processing and Management IChair: David Al-DabassCo-Chair: Zhaohao Sun

A New Hybrid Methodology for Intelligent Chinese Character RecognitionDavid Al-Dabass, David Evans, Manling Ren

Using associative classification for predicting HIV-1 drug resistanceAnantaporn Srisawat and Boonserm Kijsirikul

An Efficient Algorithm for Computing the Fitness Function of a Hydrophobic-Hydrophilic ModelMd. Tamjidul Hoque, Madhu Chetty and Laurence S Dooley

Partially Computed Fitness Function Based Genetic Algorithm for Hydrophobic-Hydrophilic ModelMd. Tamjidul Hoque, Madhu Chetty and Laurence S Dooley

Secondary Structure Prediction Using SVM and ClusteringShing H. Doong and Chi Y. Yeh

Lunch Break: 1:00 PM – 02:00 PM

Tuesday, December 07, 2:00 PM – 3:00 PM (Plenary Session - IV)

Emergent Design: Opportunities for Hybridizing Agent-based and Evolutionary ComputationUna May O'Reilly, MIT Computer Science and Artificial Intelligence Lab, USA

Tea/Coffee Break: 3:00 PM – 03:30 PM

Tuesday, December 07, 3:30 PM – 4:30 PM (Panel Discussion)

Future of Hybrid Intelligent Systems

Moderator: David Corne, Exeter University, UK

Parallel Technical Sessions: Tuesday, December 07, 04:30 PM – 06:30 PM

Session 2.3. (Room A): Anticipatory ProcessingChair: Andrew Sung

Selection of Time Series Forecasting Models based on Performance InformationPatrícia Maforte dos Santos, Teresa Bernarda Ludermir and Ricardo Bastos Cavalcante Prudêncio

A Hybrid Decision Support System Model for Disaster ManagementSohail Asghar, Damminda Alahakoon and Leonid Churilov

Polymorphic Malicious Executable Scanner by API Sequence AnalysisJ. Xu, A. H. Sung, P. Chavez, S. Mukkamala

Website Visitor Classification Using Machine LearningP. Chavez, S. Mukkamala, A. H. Sung

Innate and Acquired Immunity in Real Time SystemsDan W Taylor, David W Corne

Session 2.4. (Room B): Machine Learning and Heuristics IIChair: Witold Pedrycz

An Empirical Performance Comparison of Machine Learning Methods for Spam E-mail CategorizationChih-Chin Lai, Ming-Chi Tsai

Learning Diagnosis Profiles through Semi-Supervised Gradient Descent of Hidden Markov ModelsLaurent Jeanpierre, François Charpillet

Reconfigurable Amplifier Circuits for Adaptive Sensor Systems Employing Bio-InspirationSenthil Kumar Lakshmanan and Andreas König

Support Vector Machine and Generalized Regression Neural Network Based Classification Fusion Models forCancer Diagnosis

Muhammad Shoaib B. Sehgal, Iqbal Gondal and Laurence Dooley

Discovering Operational Signatures with Time Constraints from a Discrete Event SequencePhilippe Bouché, Marc Le Goc

Block Learning Bayesian Network Structure from DataYifeng Zeng, Kim-leng Poh

07:00 - 09:00 PM Conference Banquette

Banquette Speech: Ferment of Softcomputing in IIZUKA ConferenceSpeaker: Takeshi Yamakawa, Kyushu Institute of Technology

Lecture Concert with Japanese Traditional Musical Instruments: Koto and ShakuhachiPerformer: Tamae and Takeshi Yamakawa

Wednesday, December 08, 2004

Wednesday, December 08, 9:30 AM – 10:30 AM (Plenary Session - V)

Intelligent Robotic Systems: Safety, Security, Health and DependabilityToshio Fukuda, Nagoya University, Japan

Tea/Coffee Break: 10:30 – 11:00 AM

Parallel Technical Sessions: Wednesday, December 08, 11:00 AM – 01:00 PM

Session 3.1 (Room A): Knowledge Processing and Management IIChair: Giovanni SemeraroCo-Chair: Ickjai Lee

Downward Refinement in the ALN Description LogicNicola Fanizzi, Stefano Ferilli, Luigi Iannone, Ignazio Palmisano and Giovanni Semeraro

Hybrid Soft Categorization in Conceptual SpacesIckjai Lee

Experience Based Reasoning for Recognising Fraud and DeceptionZhaohao Sun, Gavin Finnie

An Efficient Feature Selection using Multi-Criteria in Text CategorizationDoan Son, Susumu Horiguchi

Session 3.2 (Room B): Soft Computing Systems IIChair: Gancho Vachkov

Evolutionary Artificial Neural Network Optimisation in Financial EngineeringSerge Hayward

HDGSOM: A Modified Growing Self-Organizing Map for High Dimensional Data ClusteringRasika Amarasiri, Damminda Alahakoon, Kate A. Smith

Zamin, A an Agent Based Artificial Life ModelRamin Halavati, Saeed Bagheri Shouraki, Saman Harati Zadeh, Caro Lucas, Pujan Ziaee

Growing Model Algorithm for Process Identification Based on Neural-Gas Learning and Local Linear MappingGancho Vachkov

Lunch Break: 1:00 PM – 02:00 PM

Parallel Technical Sessions: Wednesday, December 08, 02:00 PM – 03:30 PM

Session 3.3 (Room A): Features, Classification and ClusteringChair: M.Narasimha MurtyCo-Chair: Kyungmi Lee

Classification Ensembles for Shaft Test Data: Empirical Evaluation.Kyungmi Lee, Vladimir Estivill-Castro

A Fuzzy Clustering Algorithm using Cellular Learning Automata based Evolutionary AlgorithmReza Rastegar, Arash Hariri, Mohammad Reza Meybodi

Feature Selection and Classification of Gene Expression Profile in Hereditary Breast CancerMansoor Raza, Iqbal Gondal, David Green, Ross Coppel

Fuzzy Clustering with a Regularized Autoassociative Neural NetworkAlejandro Bassi, Juan D. Velásquez, Hiroshi Yasuda

Adaptive Boosting with Leader Based Learners for Classification of Large Handwritten DataT.Ravindra Babu, M.Narasimha Murty, V.K.Agrawal

Session 3.4 (Room B): Image and Signal ProcessingChair: Akira AsanoCo-Chair: Eiji Uchino

GAP Test : A Cognitive Evaluation Procedure for Shape DescriptorsAnarta Ghosh and Nicolai Petkov

A Novel Fuzzy Approach to Speech RecognitionRamin Halavati, Saeed B. Shouraki, Mahsa Eshraghi, Milad Alemzade, Pujan Ziaee

A Constructive Approach to Creating A Method for Generating ImagesKei Ohnishi, Kaori Yoshida

A New Blind Deconvolution Algorithm Based on a Gradient Method with Phase Spectral ConstraintsEiji Uchino, Noriaki Suetake and Morihiko Sakano

An Experimental Study of the Hybridization of Logistic Discriminant Analysis and Multilayer Neural Network forImage Identification

Akira Asano, Chie Muraki Asano, Koji Hotta, Megu Ohtaki, Mitsuji Muneyasu, and Takao Hinamoto

Tea/Coffee Break : 3:30 PM – 04:00 PM

Parallel Technical Sessions: Wednesday, December 08, 04:00 PM – 06:00 PM

Session 3.5 (Room A): Knowledge Processing and Management IIIChair: Kaori Yoshida

Q'tron Neural Networks for Constraint SatisfactionTai-Wen Yue, Mei-Ching Chen

Behavior Modeling Using a Hierarchical HMM ApproachShih-Yang Chiao and Costas S. Xydeas

Knowledge Discovery with SOM Networks in Financial Investment StrategySheng-Tun Li, Shu-Ching Kuo, Ming-Lung Hsu, Yi-Chung Cheng, Men-Hsieu Ho

K-Ranked Covariance Based Missing Values Estimation for Microarray Data ClassificationMuhammad Shoaib B. Sehgal, Iqbal Gondal and Laurence Dooley

A Case-Based Recommender for Task Assignment in Heterogeneous Computing SystemsSaeed Ghanbari, Mohamad Reza Meybodi, Kambiz Badie

Session 3.6 (Room B): OptimizationChair: Hisao Ishibuchi

Comparison of Local Search Implementation Schemes in Hybrid Evolutionary Multiobjective OptimizationAlgorithms

Hisao Ishibuchi and Kaname Narukawa

A Two-Phase Genetic and Set Partitioning Approach for the Vehicle Routing Problem with Time WindowsGuilherme Bastos Alvarenga, Geraldo Robson Mateus

Hierarchical Tournament Selection Genetic Algorithm for the Vehicle Routing Problem with Time WindowsGuilherme Bastos Alvarenga, Geraldo Robson Mateus

How to Deal with the VRPTW by using Multi-Agent CoalitionsImen Boudali, Wajdi Fki, Khaled Ghedira

An Hybrid Approach for a Constrained Routing ProblemJesus Fabian Lopez

Concluding Remarks : 06:00 PM

HIS’04 Plenary Abstracts

Plenary Session I:

Granular Modeling: The Synergy of Granular Computingand Fuzzy Logic

Witold PedryczDepartment of Electrical & Computer EngineeringUniversity of Alberta, Edmonton, Canada& Systems Research Institute, Polish Academy of SciencesWarsaw, Poland

Abstract: Logic is a universal language of abstract concepts of two-valued world. Fuzzy logic is acornerstone of abstraction of real (continuous) world. Granular Computing is aimed at the development,processing and exchanging information granules. Viewing such granules as abstractions of real world,their manipulation is ultimately governed by the mechanisms of logic, and fuzzy logic in particular.

In this talk, we focus on a general platform of granular modeling – a paradigm that seamlessly combinesthe concepts of information granules with their logic processing cast in the operational framework offuzzy sets. First, we outline the research agenda of granular modeling. Second, we elaborate on thearchitectural and algorithmic issues of granular models.

The talk offers a systematic view at the development of information granules (realized as fuzzy sets andfuzzy relations) through descriptive, prescriptive, and hybrid approaches. Those approaches arise underthe rubric of clustering and knowledge-based clustering. The processing core handling processinginformation granules involves a spectrum of logic constructs. Those include OR, AND, OR/AND fuzzyneurons. In conjunction to their basic logic characteristics articulated by means of “standard” logicoperators (being realized via some t- and s-norms), we show how their underlying functionality could beaugmented through more advanced constructs such as cardinal sums and uninorms. We discussseveral categories of processing units aimed at referential processing supported by matching, inclusion,dominance, and difference fuzzy neurons. In the sequel, we present fundamental topologies of logicnetworks including logic processors, fuzzy multiplexers, and referential processing units.

The issue of transparency-accuracy tradeoffs of granular models is presented along with variousmechanisms of pruning logic networks and their underlying quantification aspects (articulated in thesense of the approximation error and being viewed vis-à-vis structural complexity of the resulting logicdescription).

HIS’04 Plenary Abstracts

Plenary Session II:

Hybrid Intelligent Techniques for Intelligent Personal Assistant inDigital Convergence1

Sung-Bae ChoDepartment of Computer Science, Yonsei University,134 Shinchon-dong, Sudaemoon-ku, Seoul 120-749, Koreasbcho@cs.yonsei.ac.kr

Abstract: Recently mobile phones have become an essential tool for human communication. As morepeople use mobile phones, various services based on mobile phone networks and high-end deviceshave been developed. Smartphone which integrates the functions of personal digital assistant andmobile phone obtains world-wide reputation as new personal business assistant and entertainmentequipment because it is all-in-one device: many technologies such as wireless voice/datacommunication, digital camera, and multi-media player are converged into one device. Especially, withthe rise of the concept of ubiquitous computing, the demand for personalized intelligent service onmobile devices gets higher. However, current mobile devices have constraints of limited processingpower, and awkward interaction devices. We are in need of putting together available AI techniques tocope with these constraints and realize intelligent services in digital convergence.

There are three major issues in implementing intelligent service in constrained environment: The first isto gather information which provides meaningful features for user's state. Requiring directly explicitinformation from user can never be an intelligent technique. Niche technique should be able to providesufficient information without bothering the user and invading the user’s privacy. The second is to inferand predict user's state from collected data. Predicting user's state from data can be formulated asconventional classification task. Many AI techniques have been successfully applied to this problem. Thethird is service selection or composition. We can select one service from pre-defined service library orcompose novel services appropriate to user's state inferred dynamically.

In this talk, we will review the state-of-the-art of research efforts to develop intelligent personal assistantwith hybrid intelligent techniques, and present some prototype systems implemented in Soft computinglabs in Yonsei University based on a novel framework of hybridizing several intelligent techniques.

1 This work was supported by Brain Science and Engineering Research Program sponsored by KMOST.

HIS’04 Plenary Abstracts

Plenary Session III:

Discovering Rules of Adaptation and Interaction: From Moleculesand Gene Interaction to Brain Functions

Nikola KasabovKnowledge Engineering and Discovery Research Institute, KEDRIAuckland University of Technology, Auckland, New Zealandnkasabov@aut.ac.nz, www.kedri.info

Abstract: Many systems in Biology and Nature are characterized by a continuous adaptation and by acomplex interaction of many variables over time. Such systems can be observed at different levels of thefunctioning of a living organism, e.g.: molecular, genetic, cellular, multi-cellular, neuronal, brain function,evolution. One of the challenges for information science is to be able to represent the dynamicprocesses, to model them, and to reveal “the rules” that govern the adaptation and the variableinteraction over time.

The paper presents one approach to address the above issues through adaptive, knowledge-basedconnectionist systems. These systems evolve their structure and functionality through learning from datain both on-line and off-line incremental mode, in both supervised and unsupervised modes, and facilitatedata and knowledge integration, rule extraction and rule manipulation. Evolving Connectionist Systems(ECOS) are presented as an example of such systems. The evolving process of an ECOS is defined byparameters, “genes” [1]. ECOS extend further the classical knowledge-based neural networks [2].

Adaptive, knowledge-based connectionist systems, and ECOS in particular, are applied on data andproblems from Bioinformatics and Neuroinformatics to discover rules of adaptation and interaction. Suchproblems are: microarray gene expression analysis and profiling; gene regulatory network modeling(GRN); medical prognostic systems; modeling visual and auditory perception states of the human brain,etc [1]. Recently new, biologically plausible ECOS have been developed, called computationalneurogenetic models [3,4]. In these models the neuronal parameters correspond to real genesexpressed in the brain and related to learning processes and brain diseases [3,4]. A dynamic model of aGRN within each neuron is evolved during the modeling process that governs the neuronal processes.All neurons have a spiking behavior to form spiking neural networks, characterized by a spectral profile[3,4]. The computational neurogenetic modeling paradigm is illustrated on a case study of modelingepileptic/normal state transitions and the discovery of the gene interaction networks that are likely tocause this phenomenon. Further directions include: the development of dynamic evolving neurogeneticmodels; modeling complex brain processes and diseases; cancer prognosis; hardware implementation.

References

[1] N.Kasabov, Evolving Connectionist Systems: Methods and Applications in Bioinformatics, Brain Study andIntelligent Machines, Springer Verlag, 2002.[2] N.Kasabov, Foundations of neural networks, fuzzy systems and knowledge engineering, MIT Press, 1996.[3] N.Kasabov and L.Benuskova, Computational Neurogenetics, Journal of Computational and TheoreticalNanoscience, vol.1, No.1, American Scientific Publishers, 2004.[4] N.Kasabov, L.Benuskova, S.Wysosky, Computational neurogenetic modelling: Gene networks within neuralnetworks, Proc. IJCNN 2004, Budapest, 25-29 July, IEEE Press.

HIS’04 Plenary Abstracts

Plenary Session IV:

Emergent Design: Opportunities for Hybridizing Agent-based andEvolutionary Computation

Una May O'ReillyMIT Computer Scienceand Artificial Intelligence Labunamay@csail.mit.edu

Abstract: Recently hybridized approaches that exploit agent-based computation and evolutionarycomputation have challenged conventional assumptions about the creative power of digital design tools.A simple artificial ant system that can generate digital painterly and pencil sketching renderings startingfrom a digital photograph will be presented as one example of how design might be conceptuallyrevisited and approached as a cooperative, emergent designer-computer endeavor. A tool for architectsthat helps architects generate responsive, natural digital surfaces which can be physically fabricated willbe described. It combines evolutionary computation and generative algorithms which allows it to be bothinteractive with its user and creatively influential.

Plenary Session V:

Intelligent Robotic Systems: Safety, Security, Healthand Dependability

Toshio FukudaDept. of Micro System Engineering andDept. of Mechano-Informatics and SystemsNagoya Universityfukuda@mein.nagoya-u.ac.jp

Abstract: There have been a lot of meeds and requests for the applications of robotic system in thesociety. It is expected that intelligence will play an important role of those applications. Several topicsare introduced in intelligent system, such as safety, security and health for people with respect tomechatronics and robotics and further with the future dependable system and society.

HIS’04 Abstracts: Session 1.1

Machine Learning and Heuristics I

Margin-based Active Learning using Background Knowledge in Text MiningCatarina Silva, Bernardete Ribeiro

Text mining, also known as intelligent text analysis, text data mining or knowledge-discovery in text, refersgenerally to the process of extracting interesting and non-trivial information and knowledge from text. One of themain problems with text mining and classification systems is the lack of unlabeled data, as well as the cost oflabeling unlabeled data (Kiritchenko and Matwin 2001). Thus, there is a growing interest in exploring the use ofunlabeled data as a way to improve classification performance in text classification. The ready availability of thiskind of data in most applications makes it an appealing source of information.In this work we evaluate the benefits of introducing unlabeled data in a support vector machine automatic textclassifier. We further evaluate the possibility of learning actively and propose a method for choosing the samples tobe learned.

Hybridising Rule Induction and Multi-Objective Evolutionary Search for Optimising Water DistributionSystemsLaetitia Jourdan, David Corne, Dragan Savic, Godfrey Walters

In this article, we present our latest work with a hybrid multiobjective evolutionary algorithm called LEMMO(Learnable Evolution Model for Multi-Objective Optimization) which integrates machine learning into evolutionarysearch in a way based on Mychalski's `LEM' approach. The objective is to both improve the performance of theMOEA and to reduce the number of evaluations needed when used for optimising the design of water distributionnetworks (where evaluations are highly computationally costly). We compare LEMMO with NSGA-II and concludethat our approach is very promising for improved speed and quality in the water systems optimisation domain.

Robotic Hand-Eye Coordination: From Observation to ManipulationShahram Jafari, Ray Jarvis

In this paper, we present a new hybrid method of performing eye-to-hand coordination and manipulation toproduce a working robot named COERSU. The method is an optimized combination of two neuro-fuzzyapproaches developed by the authors: direct fuzzy servoing and fuzzy correction. The fuzzy methods are tuned byan Adaptive Neuro-Fuzzy Inference System (ANFIS). On the whole, a genetic tuner and two neuro-fuzzy networkscontribute to find the final optimum position of the robotic tooltip in order to grasp the target. Experimental resultsfrom COERSU in a table-top scenario to manipulate some soft objects (e.g. fruit) are also provided to validate themethods.

An Intelligent Model for Reconstruction of Stance Time from Faulty Gait RecordingsJoarder Kamruzzaman, Rezaul Begg

In an erroneous footfall ground-reaction force-time recording, which may occur for people with disabilities or frailelderly individuals, the stance time (ST) can be either corrupted or missing. Previous methods to estimate missingST requires force-time data from multiple force platforms and are affected by inter-step variability. This paper pre-sents a model based on Support Vector Machine (SVM) that is capable of estimating the missing ST from theavailable vertical force-timing characteris-tics with significantly high accuracy. The model was built using featurestaken from a data set of 466 sample trials of 27 subjects. A test on 40 sample trials drawn from all the subjectsrevealed an aver-age prediction accuracy of 96.63% (±2.89%). In one-fourth of the test trials, the prediction errorwas within 1.0%. The model achieves considerable im-provement over an Artificial Neural Network based modelbuilt and tested on the same data set. The effect of kernel function parameters and ε-insensitive loss function onprediction error is also analysed and presented.

Comparisons between Heuristics Based on Correlativity and Efficiency for Landmarker GenerationDaren Ler, Irena Koprinska, Sanjay Chawla

Recently, we proposed a new meta-learning approach based on landmarking. This approach, which utilises a newset of criteria for selecting landmarkers, generates a set of landmarkers that are each subsets of the candidatealgorithms being landmarked. In this paper, we experiment with three heuristics based on correlativity andefficiency. With each heuristic, the landmarkers generated using linear regression are able to estimate accuracywell, even when only utilising a small fraction of the given algorithms. The results also show that the heuristic inwhich efficiencies are estimated via 1-nearest neighbour outperformed the other heuristics.

Using Association Rules for Completing Missing DataChih-Hung Wu and Chian-Huei Wun

We present in this paper a new method for completing missing data using the concept of association rules. Thebasic idea is that association rules describe the dependency relationships among data entries in a dataset whereall data, including the missing ones, should hold the similar relationships. For a missing datum, we guess itspossible value according to related association rules. A new com-pleting procedure and a new evaluation functionare developed and presented. The evaluation function is scored according to the support, confidence, and lift ofassociation rules, which reasonably reflects the dependency relationships among existing and miss-ing data.Experimental results show that our method is feasible in completing some incomplete datasets.

HIS’04 Abstracts: Session 1.2

Web Intelligence and Data Mining

Image Clustering System on WWW using Web TextsWataru Sunayama, Akiko Nagata, and Masahiko Yachida

As the internet becomes the basic resource of information, not only texts but image retrieval systems haveappeared. However, many supply only a list of images, so we must seek desired images one by one. Imagelabeling is one solution to such a problem, in which various words are labeled to an image when extracted from asingle Web page. Therefore, this paper proposes an image clustering system that labels images by words relatedto a search keyword. Relationships are measured by Web pages in WWW. In experiments, users found theintended images faster than ordinary image search system.

Application of Machine Learning Techniques to Web-Based Intelligent Learning Diagnosis SystemChenn-Jung Huang, Ming-Chou Liu, San-Shine Chu, Chin-Lun Cheng

This work proposes an intelligent learning diagnosis system that supports a Web-based thematic learning model,which aims to cultivate learners' ability of knowledge integration by giving the learners the opportunities to selectthe learning topics that they are interested, and gain knowledge on the specific topics by surfing on the Internet tosearch related learning courseware and discussing what they have learned with their colleagues. Based on the logfiles that record the learners' past online learning behavior, an intelligent diagnosis system is used to giveappropriate learning guidance to assist the learners in improving their study behaviors and grade online classparticipation for the instructor. The achievement of the learners' final reports can also be predicted by the diagnosissystem accurately. Our experimental results reveal that the proposed learning diagnosis system can efficiently helplearn-ers to expand their knowledge while surfing in cy-berspace Web-based theme-based learning model.

Application of Support Vector Machines to Admission Control for Proportional Differentiated ServicesEnabled Internet ServersChenn-Jung Huang, Chih-Lun Cheng

A widely existing problem in contemporary web servers is the unpredictability of response time. Owing to longresponse delay, revenues of the enterprises are substantially reduced due to many aborted e-commercetransactions. Recently, re-searchers have been addressing different admission control schemes of differentiatedservice for web servers to complement the Internet differentiated services model and thereby provide QoS supportto the users of the World Wide Web. However, most of these admission control mechanisms do not guar-anteethe QoS requirements of all admitted clients under bursty workload. Although an Internet service model calledproportional differentiated service is enabled in web servers to improve the QoS guaran-tee predicament in theliterature, it still exists some impracticable assumptions and incompatible prob-lems with the current Internetprotocols. In this pa-per, we propose two algorithms for admission con-trol and traffic scheduler schemes of theweb server under proportional differentiated service, wherein a time series predictor is embedded to estimate thetraffic load of the client in the next time period. Two different approaches are used to implement the time seriespredictor due to the achievement of suc-cessful prediction in the literature. The experimen-tal results reveal thatthe proposed schemes can real-ize proportional delay differentiation service in mul-ticlass Web server effectivelywhen the support vec-tor regression is used as the time series predictor.

Intelligent Agent to Support Design in Supply Chain Based on Semantic Web ServiceIncheon Paik, Shinjiro Takami, Yuu Watanabe

In manufacture industry, better supply chain man-agement (SCM) not only improves efficiency of businessprocesses, but play important roles in a series of cost reductions. In a product manufacture, the initial design ismore important from viewpoint of the cost reduction in the whole cycle. It becomes important how a designercollects the information on necessary parts for product design, and the sys-tem to search information efficiently atdesign stage is required. Information infrastructure to support design in SCM was developed on a semantic Web

service environment that can provide interoperable service interfaces for integrated design attributes to agents andusers. Design Support Agent (DSAgent), as a client for information infrastructure, helps the designer to find thedesired product information. But this search process requires autonomy that lacks in DSAgent. As designers mustchange input values repeatedly to complete finding the desired product information, we suggest an autonomousDSAgent (ADSAgent) in this paper. To add auton-omy, we used situation calculus, which is the schema forrepresenting the dynamic changing world. The model to define the world was written in ConGolog statements,which is a logic program-ming language based on situation calculus. This will facilitate user to find the desiredproduct infor-mation and reduce user operations.

Improvement of Non-Linear Mapping Computation for Dimensionality Reduction in Data Visualization andClassificationKuncup Iswandy and Andreas König

The projection of high-dimensional data by linear or non-linear techniques is a well established technique in patternrecognition and other scientific and industrial application fields. Commonly, methods affiliated to multi-dimensional-scaling, projection pursuit or Sammon's non-linear distance preserving mapping are applied, based on gradientdescenttechniques. These suffer from well known dependence on initial or starting value and their limited ability toreach only local minimum.In this paper, stochastic search techniques are applied to the NLM to achieve lowerresidual stress or error value in competitive time.Encouraging results have been obtained for a particulardeveloped local algorithm both with regard to convergence time and residual error.

Hybrid learning scheme for Data Mining ApplicationsT.Ravindra Babu, M.Narasimha Murty, V.K.Agrawal

Classification of large datasets is a challending task in Data Mining. In the current work, we propose a novelmethod that compresses the data and classifies the test data directly in its compressed form. The work forms ahybrid learning approach integrating the activities of data abstraction, frequent item generation, compression,classification and use of rough sets.

HIS’04 Abstracts: Session 1.3

Soft Computing Systems I

Improving a Pittsburgh Learnt Fuzzy Rule Base Using Feature Subset SelectionPablo A. D. Castro, Daniel M. Santoro, Heloisa A. Camargo, Maria C. Nicoletti

This paper investigates the problem of feature subset selection as a pre-processing step to a method which learnsfuzzy rule bases using genetic algorithm (GA) implementing the Pittsburgh approach. Four feature subset selectionmethods are investigated in the context of learning fuzzy rule bases. Two of them are filter methods namely, theRelief-E and the C-Focus. The other two are wrapper methods using GA as their search process; one implementsthe instance-based method 1-NN and the other, the constructive neural network algorithm DistAl. Results of theexperiments conducted in three domains are presented and discussed; they show that methods which learn fuzzyrule bases can benefit from feature subset selection methods.

Ensemble of Linear Perceptrons with Confidence Level OutputPitoyo Hartono, Shuji Hashimoto

In this study we introduce an ensemble of neural networks, in which each member is a linear perceptron. Our mainobjective is to build an ensemble of neural networks that can automatically and effectively divide the problemspace and assign a subspace to each member. By assigning only a portion of the problem space, we expect thatthe learning difficulty for each member can be reduced, thus leading to better classification ability. To investigatethe effectiveness of the proposed method in dividing the problem space, in this paper we deal with ensembleconsists only of linear perceptrons, each with an additional output neuron that indicates the confidence level of theoutput.

First-Order Logical Neural NetworksThanupol Lerdlamnaochai and Boonserm Kijsirikul

Inductive Logic Programming (ILP) is a well known machine learning technique in learning concepts from relationaldata. Nevertheless, ILP systems are not robust enough to noisy or unseen data in real world domains.Furthermore, in multi-class problems, if the example is not matched with any learned rules, it cannot be classified.This paper presents a novel hybrid learning method to alleviate this restriction by enabling Neural Networks tohandle first-order logic programs directly. The proposed method, called First-Order Logical Neural Network(FOLNN), is based on feedforward neural networks and integrates inductive learning from examples andbackground knowledge. We also propose a method for determining the appropriate variable substitution in FOLNNlearning by using Multiple-Instance Learning (MIL). In the experiments, the proposed method has been evaluatedon two first-order learning problems, i.e., the Finite Element Mesh Design and Mutagenesis and compared with thestate-of-the-art, the PROGOL system. The experimental results show that the proposed method performs betterthan PROGOL.

Performance Studies on KBANNR.P. Jagadeesh Chandra Bose, G. Nagaraja

Hybrid models combining the analytical (rule-based) and connectionist(artificial neural network (ANN)) paradigmsare called Knowledge Based Neural Networks (KBNN). The Knowledge Based Artificial Neural Network(KBANN) isone such model that makes use of the domain theory represented as propositional rules and training examples. Inthis article we analyze the performance of KBANN and suggest some ideas of improving its capabilities. A study onthe effect of inductive bias on KBANN, use of adaptive learning algorithms instead of the standard backpropagationto improve the training times and the use of regularization methods for improving generalization performance ispresented. It is shown that for better performance, the initial weight assignment to links obtained by domain theory

varies with the domain and the performance of KBANN is improved by using regularization methods like the weightdecay along with Rprop training algorithm instead of the standard backpropagation.

Artificial Neural Networks Applied to Long-term Electricity Demand ForecastingMostafa Al Mamun, Ken Nagasaka

The electric power demand in Japan has steadily increased and the load factor of total power system hasdecreased. It is therefore very important to the utilities to have advance knowledge of their electrical load, so thatthey may ensure that this load is met and to minimize any interruptions to their service. One of the important pointsfor forecasting the long-term load in Japan is to take into account the past and present economic situations andpower demand. These points were considered in this study. The proposed Artificial Neural Network (ANN) that isRadial Basis Function Network (RBFN) has also showed that the changes in loads are a reflection of economy.Here, prediction of peak loads in Japan up to year 2015 is discussed using the RBFN and the maximum demandsfor 2001 through 2015are predicted to be elevated from 179.42GW to 209.18GW. The annual average rate of loadgrowth seen per ten years until 2015 is about 1.39%.

HIS’04 Abstracts: Session 1.4

Agent Systems, Design and Architectures I

Mobile Agent-Based Execution ModellingWenyu Qu and Hong Shen

Mobile agent-based technology has attracted considerable interest in both academia and industry in recent years.Fault tolerance isof paramount importance to integrate mobile agent-based technologyinto today's e-society. In thispaper, we propose a mobileagent-based fault-tolerant execution model and analyze thestochastic nature of mobileagents, including the transaction timefrom node to node, the life expectancy of mobile agents, and thepopulationdistribution of mobile agents. Our approach exploits anew way to design fault-tolerant mobile agent-driven system.Ouranalysis provides useful tools for effectively estimating theperformance of agent-driven systems.

Agent-based Modelling: A Case Study in HIV EpidemicEyob Teweldemedhin, Tshilidzi Marwala, Conrad Mueller

This research presents an agent-based, bottom-up modelling approach to develop a simulation tool for estimatingand predicting the spread of the Human Immunodeficiency Virus (HIV) in a given population. HIV is mainly asexually transmitted disease (STD) causing a serious problem to human health. The virus is transmitted from aninfected person to another who was previously healthy through different biological, social and environmentalfactors. The research develops the simulation tool by modelling these factors by agents. Although research hasand is being conducted to estimate and predict the spread of the HIV epidemic, the proposed research seeks toinvestigate the spread using a different approach. The previous models used a top-down modelling approach.They are built from the general characteristics and behaviours of the population. They have not explored thepotential use of agent technology. This research attempts to investigate the flexibility that the multi-agent systemoffers. Agent-based models are close to the situations that exist in a given real system that consists of autonomouscomponents interacting with each other. The modelling approach has the advantage of observing the interactionmade between agents, which is a difficult task in the top-down modelling approach. The research investigates theperformance of the tool and presents the first results obtained.

Cooperative Game Theory within Multi-Agent Systems for Systems SchedulingDerek Messie, Jae C. Oh

Research concerning organization and coordination within multi-agent systems continues to draw from a variety ofarchitectures and methodologies. The work presented in this paper combines techniques from game theory andmulti-agent systems to produce self-organizing, polymorphic, lightweight, embedded agents for systemsscheduling within a large-scale real-time systems environment. Results show how this approach is used toexperimentally produce optimum real-time scheduling through the emergent behavior of thousands of agents.These results are obtained using a SWARM simulation of systems scheduling within a High Energy Physicsexperiment consisting of 2500 digital signal processors.

A Rationality-based Modeling for Coalition SupportJae C. Oh, Nathaniel Gemelli and Robert Wright

We present a game theoretic model for multi-agent resource distribution and allocation where agents in theenvironment must help each other to survive. Each agent maintains a set of two-tuples F = (A, P) called "friendshipvalues" representing "actual friendship" and "perceived friendship". The model directly addresses problems inreputation management schemes in multi-agent systems and Peer-to-Peer distributed systems. We presentalgorithms for maintaining the friendship values as well as a utility equation used in each agent's decision making.For an application problem, we adapted our formal model to the military coalition support problem in peace-

keeping missions. Simulation results show that efficient resource allocation and sharing with minimumcommunication cost is achieved without centralized control.

HIS’04 Abstracts: Session 2.1

Agent Systems, Design and Architectures II

Integrative Early Requirements Analysis for Agent-Based SystemsLongbing Cao, Chengqi Zhang, Dan Luo, Wanli Chen, Neda Zamani

Goal-oriented requirements analysis is promising for agent-oriented requirement engineering. In this paper, wepresent an integrative modeling framework for agent-oriented early requirements analysis; this frameworkimplements goal-oriented requirement analysis. The integrative modeling combines visual modeling and formalmodeling together. Extended i* framework is used for building visual models; formal specifications are defined byfirst-order linear-time temporal logic. Both visual and formal models are outlined through a practical agent-basedsystem F-TRADE . The integrative modelling seems to model early requirements completely and precisely, andbenefit refinement and conflict management in building agent systems.

Motivation Based Behavior in Hybrid Intelligent Agents for Intention Reconsideration Process in VesselBerthing ApplicationsPrasanna Lokuge and Damminda Alahakoon

Strategic planning and dynamism in decision making are essential factors in a vessel berthing application in anycontainer port to assure faster turnaround time and high productivity. BDI agents have been used in manyapplications with limited capabilities. In this paper, we propose a new hybrid BDI architecture with learningcapacities overcoming some limitations exists in the generic BDI agent model. A new “knowledge AcquisitionModel” ( KAM) module is proposed with a supervised neural network and Adaptive neuro fuzzy inference system(ANFIS) in the intention reconsideration process of the agent model. Commitment strategy of the new intentionreconsideration process is based on the motivation of the state transitions and the effect of belief changes in theenvironment.

Reinforcement Learning_Hierarchical Neuro-Fuzzy Politree Model for Control of Autonomous AgentsKarla Figueiredo, Marley Vellasco, Marco Pacheco, Flávio Souza

This work presents a new hybrid neuro-fuzzy model for automatic learning of actions taken by agents. The mainobjective of this new model is to provide an agent with intelligence, making it capable, by interacting with itsenvironment, to acquire and retain knowledge for reasoning (infer an action). This new model, namedReinforcement Learning Hierarchical Neuro-Fuzzy Politree (RL-HNFP), descends from the Reinforcement LearingHierarchical Neuro-Fuzzy BSP (RL-HNFB) that uses Binary Space Partitioning. By using hierarchical partitioningmethods, together with the Reinforcement Learning (RL) methodology, a new class of Neuro-Fuzzy Systems(SNF) was obtained, which executes, in addition to automatically learning its strucutre, the autonomous learning ofthe actions to be taken by an agent. These characteristics represent an important differential when compared withthe existing intelligent agents learning systems. The obtained results demonstrate the potential of this new model,which operates without any prior information, such as number of rules, rules specification, or number of partitionsthat the input space should have.

An Image-Guided Heuristic for Planning an Exhaustive EnumerationJoselito J. Chua and Asha Rao

This paper proposes a heuristic for planning a distributed search that yields an exhaustive enumeration. The paperis motivated by the need to search for binary extremal self-dual codes. The proposed technique represents thesearch space as an image. A partial image is obtained by a preliminary sampling of the search space. We proposean image restoration algorithm which can be applied on the partial image in order to identify regions of interest, andprioritize the search accordingly. Experimental results show that the technique can guide the search effectively.

A Novel Adaptive Decision Making Agent Architecture Inspired by Human Behavior and Brain StudyModelsL. K. Wickramasinghe, L.D. Alahakoon

Intelligent agent technology, which expects to combine the marked trends in history of computing such as ubiquity,interconnection, intelligence, delegation and human orientation can be considered as a step towards the next stageof artificial intelligence. This new technology attempts to reduce the gap between man and machine. Theremarkable ability of a human being to make decisions is an ongoing learning and evolutionary process. There-fore, when reducing the man-machine gap, one of the main issues to address is how to make agents decisionmakers in a human oriented way. The paper presents novel conceptual agent framework to provide human likedecisions inspired by human behavior and brain study models. The proposed learning and evolutionary agentarchitecture make the agent capable of handling the dynamism in the environment too. The experiments illustratedwith the banking application demonstrate how the proposed framework enables a software agent to make deci-sions in a human oriented manner.

HIS’04 Abstracts: Session 2.2

Knowledge Processing and Management I

A New Hybrid Methodology for Intelligent Chinese Character RecognitionDavid Al-Dabass, David Evans, Manling Ren

A new methodology is proposed to combine fuzzy possibilistic reasoning with knowledge mining of syntaxdynamics using hybrid recurrent nets. The structure of Chinese characters consist of a 3-layer hierarchy ofcharacter, radical and stroke. Fuzzy possibilistic reasoning is an appropriate set of algorithmic tools to aidautomatic recognition of these characters. Associative memory artificial neural network algorithms form a suitabletechnique for realizing these concepts. In implementing these techniques several issues are explored: vaguenessof radicals, their situation, position invariance, extraction order and shape. Hybrid recurrent nets are then proposedto deal with recognition at the syntax level. Each character is the 2-D output of a syntax generating system which issubjected to a knowledge mining process to determine its behavior parameters. The knowledge miningarchitecture consists of an extensible recurrent hybrid net hierarchy of multi-agents where the composite behaviorof agents at any one level is determined by those of the level immediately below. Extensive results are obtained todemonstrate the quality of the algorithms in dealing with the range of difficulties inherent in the problem.

Using associative classification for predicting HIV-1 drug resistanceAnantaporn Srisawat and Boonserm Kijsirikul

Drug resistance testing: genotyping and phenotyping is an important role in management of HIV-1 infections. Toovercome drawbacks of genotyping and phenotyping to predict the HIV-1 drug resistance, predicting of phenotypicresistance from genotypic data is an interesting task. In this paper, CBA algorithm was used to discover therelationship between the amino acid positions and drug susceptibility and to construct the classifiers to predictphenotypic resistance for 6 protease inhibitors. The performance of the prediction was measured by 10-fold cross-validation. The best model provided the accuracy between 84.11% and 92.64% for all 6 protease inhibitors. Inaddition, the average accuracy of 6 drugs of the prediction using CBA algorithm provided the best performancewhen compared with HIVdb, SVM, and REG algorithms.

An Efficient Algorithm for Computing the Fitness Function of a Hydrophobic-Hydrophilic ModelMd. Tamjidul Hoque, Madhu Chetty and Laurence S Dooley

The protein folding problem is a minimization problem in which the energy function is often regarded as the fitnessfunction. There are several models for protein folding prediction including the Hydrophobic-Hydrophilic (HP) model.Though this model is an elementary one, it is widely used as a test-bed for faster execution of new algorithms.Fitness computation is one of the major computational parts of the HP model. This paper proposes an efficientsearch (ES) approach for computing the fitness value requiring only O(n) complexity in contrast to the full search(FS) approach that requires O(n2) complexity . The efficiency of the proposed ES approach results due to itsutilization of some inherent properties of the HP model. The ES approach represents residues in a cartesiancoordinate framework and then uses relative distance and coordinate polarity to reduce complexity.

Partially Computed Fitness Function Based Genetic Algorithm for Hydrophobic-Hydrophilic ModelMd. Tamjidul Hoque, Madhu Chetty and Laurence S Dooley

Fitness computation after each crossover or mutation operation in Genetic Algorithm (GA) requires computationaltime that increases with the increasing length of the chromosome. In this paper, an efficient GA is proposed forprotein folding prediction based on the Hydrophobic-Hydrophilic (HP) model. The partial fitness of the parentcomputed from one end of sequence till crossover or mutation point is utilized for the computation of the fitness ofthe child. The calculated value of the partial fitness is stored with the corresponding chromosome. Although theapproach requires additional memory for each hydrophobic residue of each chromosome, the computation time isreduced significantly which is more important than the memory overhead.

Secondary Structure Prediction Using SVM and ClusteringShing H. Doong and Chi Y. Yeh

Protein secondary structure can be used to help determine the tertiary structure via the fold recognition method.Predicting the secondary structure from the protein sequence has attracted the attention of many researchers.Support Vector Machine (SVM) is a new learning algorithm that has been successfully applied to many predictionproblems. However, the algorithm takes a long time to train the prediction model when a large data set is present.It becomes important to revise the method so that the time performance is improved while the accuracyperformance is maintained. In this study, we implement a genetic algorithm to cluster the training set before aprediction model is built. Using position specific scoring matrix (PSSM) as part of the input, the hybrid methodachieves good performances on sets of 513 non-redundant protein sequences and 294 partially redundantsequences. The results also show that clustering achieves the goal of data preprocessing differently on redundantand non-redundant sets, and it seems almost preferable to cluster the data before prediction is preformed.

HIS’04 Abstracts: Session 2.3

Anticipatory Processing

Selection of Time Series Forecasting Models based on Performance InformationPatrícia Maforte dos Santos, Teresa Bernarda Ludermir and Ricardo Bastos Cavalcante Prudêncio

In this work, we proposed to use the Zoomed Ranking approach to rank and select time series models. ZoomedRanking, originally proposed to generate a ranking of candidate algorithms, is employed to solve a givenclassification problem based on performance information from previous problems. The problem of model selectionin Zoomed Ranking was solved in two distinct phases. In the first phase, we selected a subset of problems fromthe instances base that were similar to the new problem at hand. This selection is made using the k-NearestNeighbor algorithm, whose distance function uses the characteristics of the series. In the second phase, theranking of candidate models was generated based on performance information (accuracy and execution time) ofthe models in the series selected from the previous phase. Our experiments using the Zoomed Ranking revealedencouraging results.

A Hybrid Decision Support System Model for Disaster ManagementSohail Asghar, Damminda Alahakoon and Leonid Churilov

Model integration is one of the most important and widely researched areas in model management of DecisionSupport Systems (DSS). In disaster management area, independent DSS models handle specific decision-makingneeds, but it is possible that the need for combination of these models will be required. Therefore, the need formodel inte-gration and selection of such models arises. This paper presents the idea of decision support modelintegration based on software agents in an interac-tive disaster management domain. In this environ-ment anautomated model agent communicates with the other decision support system models and pre-sents the hybriddecision support system model as a solution. This system starts with minimal informa-tion about the user‚spreferences, and preferences are elicited and inferred incrementally by analyzing the needs and requirements ofthe user. The objec-tive of the system is to present a hybrid DSS model, which combines the functionality ofnumber of dif-ferent existing models. This system contrasts with traditional decision support systems where amodel is elicited beforehand or is constructed by human experts.

Polymorphic Malicious Executable Scanner by API Sequence AnalysisJ. Xu, A. H. Sung, P. Chavez, S. Mukkamala

The proliferation of viruses, Trojans, and other malicious code in recent years has presented a serious thread toboth organizations and personals. Polymorphic computer viruses are the most complex and difficult maliciouscodes to detect, often requiring anti-virus companies to spend days or months creating the routines needed tocatch a single polymorphic. In this paper, we propose a new approach for detection of polymorphic viruses inWindows platform. Our approach rests on an analysis based on a Windows API calling sequence that reflects thebehavior of a piece of particular code. The analysis is carried out directly on Portable Executable (PE) binary code.It is achieved in two major steps: construction API calling sequences for both known virus and suspicious code,and similarity measurement between the two sequences.

Website Visitor Classification Using Machine LearningP. Chavez, S. Mukkamala, A. H. Sung

Classifying website visitors allows organizations to present customized content and effectively allocate resources.Traditional methods of visitor classification involve tracking individual users over many sessions via a uniqueidentifier such as the IP address or a cookie. These methods are either too general or strip the visitor of a level ofprivacy. In this paper we use machine learning techniques to classify visitors of a data-centric website using aminimal amount of information and without a unique identifier. We are able to group visitors into groups withoutextended user tracking.

Innate and Acquired Immunity in Real Time SystemsDan W Taylor, David W Corne

In most potential industrial applications of artificial immune systems for early fault detection, some form of simplefault detection system already exists. We propose that this existing layer of simple, generally rule-based faultdetection is analogous to innate immunity in the natural immune system. We argue that the artificial acquiredimmune system should focus on the detection of fault conditions not already covered by the innate immunesystem, most importantly on the very early symptoms of faults which, we believe, are often very similar to self.This has implications for detector generation algorithms. We test two novel detector generation algorithms whichaddress this issue, using temperature data from refrigerated cabinets in UK supermarkets. Results show that thelocation of detectors within problem space is very important and that detector sets concentrated close to self inproblem space are better at detecting the early stages of fault.

HIS’04 Abstracts: Session 2.4

Machine Learning and Heuristics II

An Empirical Performance Comparison of Machine Learning Methods for Spam E-mail CategorizationChih-Chin Lai, Ming-Chi Tsai

The increasing volume of unsolicited bulk e-mail (also known as spam) has generated a need for reliable anti-spam filters. Using a classifier based on machine learning techniques to automatically filter out spam e-mail hasdrawn many researchers' attention. In this paper, we review some of relevant ideas and do a set of systematicexperiments on e-mail categorization, which has been conducted with four machine learning algorithms applied todifferent parts of e-mail t. Experimental results reveal that the header of e-mail provides very useful information forall the machine learning algorithms considered to detect spam e-mail.

Learning Diagnosis Profiles through Semi-Supervised Gradient Descent of Hidden Markov ModelsLaurent Jeanpierre, François Charpillet

In this paper, we consider the problem of adapting the model of a diagnosis -helping module which interacts withhuman experts. The approach consists of enforcing strong semantics in the model, so that this interaction may beas intuitive as possible. When learning the model, the problem consists in respecting these semantics whilelearning with few data. We addressed this problem through a semi-supervised gradient descent algorithm appliedto partially observable Markov models with fuzzy observations. This method optimizes several criteria at once,guiding the search to a compromise between the Expert's directives and objective evaluations. This method hasbeen successfully applied to a telemedicine application where the system monitors dialyzed patients and alertsnephrologists.

Reconfigurable Amplifier Circuits for Adaptive Sensor Systems Employing Bio-InspirationSenthil Kumar Lakshmanan and Andreas König

In particular primary electronics of sensor systems is strongly subject to deviations and degradations caused byenvironmental and manufacturing conditions. Current approaches cope with these challenges by calibration ortrimming techniques. More recent approaches from the field of evolutionary electronics offer considerableextensions, incorporating also issues of fault-tolerance and self-repair. State-of-the-art evolvable analog hardwarebases on field-programmable-transistor-arrays (FPTA) and start from primal soup for each new problem. Our workdeals with the crucial issue to efficiently include the wealth of existing engineering design knowledge into theotherwise attractive concept. For the practically relevant task of sensor amplifiers, a particular flexible FPTAarchitecture is developed, verified and physically implemented in a 0.35 mm CMOS technology.

Support Vector Machine and Generalized Regression Neural Network Based Classification Fusion Modelsfor Cancer DiagnosisMuhammad Shoaib B. Sehgal, Iqbal Gondal and Laurence Dooley

This paper presents decision-based fusion models to classify BRCA1, BRCA2 and Sporadic genetic mutations forbreast and ovarian cancer. Different ensembles of base classifiers using the stacked generalization technique havebeen proposed including Support Vector Machines (SVM) with linear, polynomial and radial base function kernels.A Generalized Regression Neural Networks (GRNN) is then applied to predict the mutation type based on theoutputs of base classifiers, and experimental results will show that the new proposed fusion methodology forselecting the best and removing weak classifiers outperforms single classification models.

Discovering Operational Signatures with Time Constraints from a Discrete Event SequencePhilippe Bouché, Marc Le Goc

This paper aims at showing a method to discover signatures (or models of chronicles) from a discrete eventsequence (alarms) generated by a Monitoring Cognitive Agent (MCA). When the counting process of the events

generated by a couple (Process, MCA) behaves like a Poisson process, this couple can be considered asstochastic discrete event generator SDEG(Pr, MCA) and modeled as a superposition of Poisson and anhomogeneous discrete time Markov chain. The 'BJT' algorithm uses these two representations in order to help inthe discovering of signatures. The results obtained on an industrial process monitored with a Sachem system havebeen validated by Experts, confirming so the relevance of the approach within an industrial frame.

Block Learning Bayesian Network Structure from DataYifeng Zeng, Kim-leng Poh

Existing methods for learning Bayesian network structures run into the computational and statistical problemsbecause of the following two reasons: a large number of variables and a small sample size for enormous variables.Adopting the divide and conquer strategies, we propose a novel algorithm to learn Bayesian network structures.The method partitions the variables into several blocks that are overlapped with each other. The blocks are learnedindividually with some constraints obtained from the learned overlap structures. After that, the whole network isrecovered by combining the learned blocks. Comparing with some typical learning algorithms on golden Bayesiannetworks, our proposed methods are efficient and effective. It shows a large potential capability to be scaled up.

HIS’04 Abstracts: Session 3.1

Knowledge Processing and Management II

Downward Refinement in the ALN Description LogicNicola Fanizzi, Stefano Ferilli, Luigi Iannone, Ignazio Palmisano and Giovanni Semeraro

We focus on the problem of specialization in a Description Logics (DL) representation, specifically the ALNlanguage. Standard approaches to learning in these representations are based on bottom-up algorithms thatemploy the Least Common Subsumer (lcs) operator, which, in turn, produces overly specific (overfitting) and stillredundant concept definitions. In the dual (top-down) perspective, this issue can be tackled by means of an ILPdownward operator. Indeed, using a mapping from DL descriptions onto a clausal representation, we define aspecialization operator, computing maximal specializations of a concept description on the grounds of the availablepositive and negative examples.

Hybrid Soft Categorization in Conceptual SpacesIckjai Lee

Understanding the process of categorization is of great importance for building intelligent agents. Formulatedcategories help agents find information easier and understand the external world better. Instance-basedcategorization and prototype-based categorization have been two dominant approaches in the AI community.However, they share some drawbacks in common. First, they are crisp boundary-based hard categorizations(similar to classification). Second, they are not well-suited for dynamic category learning and formation. In thispaper, we propose a hybrid soft categorization in the conceptual level that overcomes these drawbacks. The hybridsoft categorization merges the two popular hard categorizations and provides a robust fuzzy boundary-based softcategorization.

Experience Based Reasoning for Recognising Fraud and DeceptionZhaohao Sun, Gavin Finnie

Fraud, deception and their recognition have received increasing attention in multiagent systems (MAS), e-commerce, and agent societies. However, little attention has been given to the theoretical foundation for fraud anddeception from a logical viewpoint. This paper will fill this gap by arguing that experience-based reasoning (EBR) isa logical foundation for recognising fraud and deception. It provides a logical analysis of deception which classifiesrecognition of deception into knowledge-based deception recognition, inference-based deception recognition, andhybrid deception recognition. It will examine the relationship between EBR and fraud as well as deception. It usesEBR to recognize fraud and deception in e-commerce and MAS. The proposed approach will facilitate researchand development of recognition of fraud and deception in e-commerce.

An Efficient Feature Selection using Multi-Criteria in Text CategorizationDoan Son, Susumu Horiguchi

This paper considers the problem of feature selection in text categorization. A new approach of feature selectionbased on multi-criteria ranking of features is propsed. Based on a threshold value for each criterion, a newprocedure for feature selection is proposed and applied to a text categorization. Experiments dealing with theReuters-21578 benchmark data and the naive Bayes algorithm show that the proposed approach outperformsperformances in compare to conventional feature selection methods.

HIS’04 Abstracts: Session 3.2

Soft Computing Systems II

Evolutionary Artificial Neural Network Optimisation in Financial EngineeringSerge Hayward

Analytical examination of loss functions’ families demonstrates that investors' utility maximisation is determined bytheir risk attitude. In computational settings, stock traders’ fitness is assessed in response to a slow-step increasein the value of the risk aversion coefficient. The experiment rejects the claims that the accuracy of the forecastdoes not depend upon which error-criteria are used and none of them is related to the profitability of the forecast.Profitability of networks trained with L6 loss function appeared to be statistically significant and stable, althoughlinks between loss functions and accuracy of forecasts were less conclusive.

HDGSOM: A Modified Growing Self-Organizing Map for High Dimensional Data ClusteringRasika Amarasiri, Damminda Alahakoon, Kate A. Smith

The Growing Self Organizing Map (GSOM) algorithm is a variant of the Self Organizing Map (SOM). It has adynamically growing structure that adapts to the natural structure of the data. It has been identified that the growingof the GSOM can get negatively affected when used with very large dimensional data such as those in text andDNA data sets. This paper addresses these issues and presents a modified version of the GSOM called the HighDimensional GSOM (HDGSOM). The algorithm and experimental results showing the improved performance of theHDGSOM are also presented.

Zamin, A an Agent Based Artificial Life ModelRamin Halavati, Saeed Bagheri Shouraki, Saman Harati Zadeh, Caro Lucas, Pujan Ziaee

Zamin artificial life model is designed to be a general purpose environment for researches on evolution of learningmethods, living strategies and complex behaviors and is used in several studies thus far. As a main target forZamin's design has been its expandability and ease of problem definition, a new agent based structure for thisartificial world is introduced in this paper, which is believed to be much easier to use and extend. In this newmodel, all control and world running processes are done by agents. Therefore, any change in world processesdoes not require recoding the main engine and can be done just by altering the behavior of one or some agents.To have an easier interface for the design of new organisms, all creatures' communications with the world level isdone through a common message map, thus, a designer just needs to code the required parts and append them tothe main system to process the necessary messages. And at last, extending the model can be done with muchless effort, as it can be done easily by creating new agents that handle the new tasks. This model is implementedand some coding differences with previous model are presented.

Growing Model Algorithm for Process Identification Based on Neural-Gas Learning and LocalLinear MappingGancho Vachkov

The paper proposes a growing type of identification model, based on Local Linear Mapping. Feedback informationabout the approximation error of the current model is used in order to make decision for insertion of new neuronsin the area with the biggest evaluation error. It is shown in the paper that the final produced model has a betterapproximation and generalization ability than some other known learning algorithms.

HIS’04 Abstracts: Session 3.3

Features, Classification and Clustering

Classification Ensembles for Shaft Test Data: Empirical Evaluation.Kyungmi Lee, Vladimir Estivill-Castro

A-scans from ultrasonic testing of long shafts are complex signals. The discrimination of different types of echoesis of importance for non-destructive testing and A-scans from ultrasonic testing of long shafts are complex signals.The discrimination of different types of echoes is of importance for non-destructive testing and equipmentmaintenance. Research has focused on selecting features of physical significance or exploring classifier likeArtificial Neural Networks and Support Vector Machines. This paper confirms the observation that there seems tobe uncorrelated errors among the variants explored in the past, and therefore an ensemble of classifiers is toachieve better discrimination accuracy. We explore the diverse possibilities of heterogeneous and homogeneousensembles, combination techniques, feature extraction methods and classifiers types and determine guidelines forheterogeneous combinations that result in superior performance.

A Fuzzy Clustering Algorithm using Cellular Learning Automata based Evolutionary AlgorithmReza Rastegar, Arash Hariri, Mohammad Reza Meybodi

In this paper, a new fuzzy clustering algorithm that uses cellular learning automata based evolutionary computing(CLA-EC) is proposed. The CLA-EC is a model obtained by combining the concepts of cellular learning automataand evolutionary algorithms. The CLA-EC is used to search for cluster centers in such a way that minimizes theclustering criterion. The simulation results indicate that the proposed algorithm produces clusters with acceptablequality with respect to clustering criterion and provides a performance that is superior to that of the C-meansalgorithm.

Feature Selection and Classification of Gene Expression Profile in Hereditary Breast CancerMansoor Raza, Iqbal Gondal, David Green, Ross Coppel

Correct classification and prediction of tumor cells are essential for successful diagnosis and reliable futuretreatment. However, it is very challenging to distinguish between tumor classes using thousands of geneexpression data of microar-ray. Removing irrelevant genes is very helpful for us to learn the relationship betweengenes and tu-mors. In this paper we have used two methods: Multivariate Permutation Test (MPT) and Signifi-cantAnalysis of Microarray (SAM) to select sig-nificant genes for feature selection. Using those selected features, weapplied Support Vector Ma-chine, (SVM) with polynomial, radial and linear kernels, to predict the class of testingdata. Our re-sult shows that all the samples are classified cor-rectly. We have achieved 100% accuracy in classifi-cation among all the samples with polynomial ker-nel of SVM while linear kernel shows no misclassi-ficationamong BRCA1-BRCA2 and BRCA1-sporadic.

Fuzzy Clustering with a Regularized Autoassociative Neural NetworkAlejandro Bassi, Juan D. Velásquez, Hiroshi Yasuda

We propose a fuzzy clustering method that relies on an artificial neural network scheme based on an encoder-decoder architecture with autoassociative training. The encoder is designed to implement a set of competing fuzzymembership functions which are trained to fit the data so that the decoder reconstruction error is minimized. Inorder to enforce a suitable cluster partitioning and membership distribution, the critical factor of the method is anentropy based regularization that constrains the encoder outputs. We present the results of our approach appliedto synthetic data sets featuring both disjoin and intersecting compact clusters.

Adaptive Boosting with Leader Based Learners for Classification of Large Handwritten DataT.Ravindra Babu, M.Narasimha Murty, V.K.Agrawal

Boosting is a general method for improving the accuracy of a learning algorithm. AdaBoost, short form for AdaptiveBoosting method, consists of repeated use of a weak or a base learning algorithm to find corresponding weakhypothesis by adapting to the error rates of the individual weak hypotheses. A large, complex handwritten data isunder study. A repeated use of weak learner on the huge data results in large amount of processing time. In viewof this, instead of using the entire training data for learning, we propose to use only prototypes. Further, in thecurrent work, the base learner consists of a nearest neighbour classifier that employs prototypes generated using"leader" clustering algorithm. The leader algorithm is a single pass algorithm and is linear in terms of time as wellas computation complexity. The prototype set alone is used as training data. In the process of developing analgorithm, domain knowledge of the Handwritten data, which is under study, is made use of. With the fusion ofclustering, prototype selection, AdaBoost and Nearest Neighbour classifier, a very high classification accuracy,which is better than reported earlier on the considered data, is obtained in less number of iterations. The procedureintegrates clustering outcome in terms of prototypes with boosting.

HIS’04 Abstracts: Session 3.4

Image and Signal Processing

GAP Test : A Cognitive Evaluation Procedure for Shape DescriptorsAnarta Ghosh and Nicolai Petkov

With inspiration from psychophysical researches of the human visual system we propose a novel aspect and amethod for performance evaluation of contour based shape recognition algorithms regarding their robustness toincompleteness of contours. We use complete contour representations of objects as a reference (training) set.Incomplete contour representations of the same objects are used as a test set. The amount of incompleteness isquantified using the percentage of contour pixels retained. The performance of an algorithm is reported using therecognition rate as a function of the degree of incompleteness. We consider three types of incomplete contourrepresentations, viz. segment-wise deletion, occlusion and random pixel depletion. As an illustration, therobustness of two shape recognition algorithms to contour incompleteness is evaluated. These algorithms useshape context and distance multiset as local shape descriptors. Qualitatively, both algorithms mimic human visualperception in the sense that recognition performance monotonously increases with the degree of completenessand that they perform best in the case of random depletion and worst in the case of occluded contours. Thedistance multiset based method performs better than the shape context based method in this test framework.

A Novel Fuzzy Approach to Speech RecognitionRamin Halavati, Saeed B. Shouraki, Mahsa Eshraghi, Milad Alemzade, Pujan Ziaee

This paper presents a novel approach to speech recognition using fuzzy modeling. The task begins withconversion of speech spectrogram into a linguistic description based on arbitrary colors and lengths. Whilephonemes are also described using these fuzzy measures, and recognition is done by normal fuzzy reasoning, agenetic algorithm optimizes phoneme definitions so that to classify samples into correct phonemes. The method istested over a standard speech data base and the results are presented.

A Constructive Approach to Creating A Method for Generating ImagesKei Ohnishi, Kaori Yoshida

This paper proposes a constructive approach to creating a method for generating images, which acquiresmechanisms to draw images that users provide. The constructive approach relies on a combination of productionrules and an optimization technique. The method realized based on the concept of the constructive approachcombines simple production rules for cellular automata with a genetic algorithm. The realized method is applied tocreating methods for generating simple binary images, and the results shows that the proposed approach iscapable of acquiring mechanisms to draw given images.

A New Blind Deconvolution Algorithm Based on a Gradient Method with Phase Spectral ConstraintsEiji Uchino, Noriaki Suetake and Morihiko Sakano

A new blind deconvolution method with additional phase spectralconstraints for the blurred image is discusssed inthis paper. The blurred image is a convolution of an original image and a point-spread function (PSF). Theproposed method is based on the following iterative procedures, i.e., a projection onto a frequency space satisfyingphase spectrum constraints, a minimization of a cost function and a projection onto an image space satisfying thesupport and nonnegativity constraints. This method restores the original image and the PSF stably with highaccuracy.

An Experimental Study of the Hybridization of Logistic Discriminant Analysis and Multilayer NeuralNetwork for Image IdentificationAkira Asano, Chie Muraki Asano, Koji Hotta, Megu Ohtaki, Mitsuji Muneyasu, and Takao Hinamoto

A hybridized classification system of the logistic discriminant analysis and the three-layer neural network isproposed. This system is basically a linear discrimination and is assisted by the neural network only for the casesthat are difficult to be classified by linear methods. This system presents a simple discrimination structure given bylinear methods, and its computational cost is much lower than the exclusive use of neural network while themisclassification rate is as low as the neural network. The ability of this system is shown experimentally in the caseof applying it to image identification problems. The computation time for the learning process is reduced to one-fifthby this method in this experiment, while the misclassification rate is remained almost the same.

HIS’04 Abstracts: Session 3.5

Knowledge Processing and Management III

Q'tron Neural Networks for Constraint SatisfactionTai-Wen Yue, Mei-Ching Chen

This paper proposes the methods to solve the constraint satisfaction problems (CSPs) using Q'tron neuralnetworks (NNs). A Q'tron NN is local-minima free if it is built as a known-energy system and is incorporated withthe proposed persistent noise-injection mechanism. The so-built Q'tron NN, as a result, will settle down if and onlyif a feasible solution is found. Additionally, such a Q'tron NN is intrinsically auto-reversible. This renders the NNoperable in a question-answering mode for extracting interested information. A concrete example, i.e., to solve theN-queen problem, will be demonstrated to highlight the main concept.

Behavior Modeling Using a Hierarchical HMM ApproachShih-Yang Chiao and Costas S. Xydeas

This paper introduces a new methodology for the hierarchical modeling of the behavior-with-time of playersoperating and interacting within a certain application domain. Behavior modeling and characterization areperformed on-line, given that a number of observations are made or sensed at regular time intervals with respect toeach player.A key element of this hierarchical behaviour modelling system architecture is a new formulation of multiple HiddenMarkov Models (HMM) with Discrete Densities operating in parallel, with each HMM accepting a single feature-related observation sequence. However the proposed classification approach recognizes the existence of possibledependencies between the observation sequences of the features obtained for a given player. This property iseffectively exploited in a new Dependent- Multi-HMM with Discrete densities (DM-HMM-D) classification approach.The proposed methodology is applied in modelling the behaviour of aircrafts operating in relatively simple 3-D "air-patrol" situations. Computer simulation results demonstrate the significant gains that can be obtained in systemclassification and modelling performance when compared to those obtained while using conventional Independent-Multi-Discrete Hidden Markov Model (IM-HMM-D) schemes.

Knowledge Discovery with SOM Networks in Financial Investment StrategySheng-Tun Li, Shu-Ching Kuo, Ming-Lung Hsu, Yi-Chung Cheng, Men-Hsieu Ho

Recently, the recession of the global economy induced the coming of a new era of low interest-rates, whichresulted in the stock market as an alternative investment channel for investors. The diversity and complication ofdomain knowledge existing in the stock market enhance its importance for developing a decision support systemwhich can gather real-time pricing information for supporting decision-making in financial investment. In this study,we tackle these challenges by proposing an integrated solution on the basis of K-chart analysis and the over-whelming self-organizing map neural networks. We not only endeavor to improve the accuracy of uncoveringtrading signals, but also to maximize the profits of trading. The resulting decision model can help investmentdecision-makers of national stable funds make the most profitable decisions. In addition, financial experts canbenefit from the ability of verifying or refining their tacit investment knowledge offered by the uncovered knowledge.

K-Ranked Covariance Based Missing Values Estimation for Microarray Data ClassificationMuhammad Shoaib B. Sehgal, Iqbal Gondal and Laurence Dooley

Microarray data often contains multiple missing genetic expression values that degrade the performance ofstatistical and machine learning algorithms. This paper presents a K ranked diagonal covariance-based missingvalue estimation algorithm (KRCOV) that has demonstrated significantly superior performance compared to themore commonly used K-nearest neighbour (KNN) imputation algorithm when it is applied to estimate missingvalues of BRCA1, BRCA2 and Sporadic genetic mutation samples present in ovarian cancer. Experimental resultsconfirm KRCOV outperformed both KNN and zero imputation techniques in terms of their classification accuracieswhen used to impute randomly missing values from 1% to 5%. The classifier used for this purpose was the

Generalized Regression Neural Network. The paper also provides a hypothesis for why KRCOV performs betterthan KNN not only for bioinformatics data but also for other data types having strong correlated values.

A Case-Based Recommender for Task Assignment in Heterogeneous Computing SystemsSaeed Ghanbari, Mohamad Reza Meybodi, Kambiz Badie

Case-based reasoning (CBR) is a knowledge-based problem-solving technique, which is based on reuse ofprevious experiences. In this paper we propose a new model for static task assignment in heterogeneouscomputing system. The proposed model is a combination of the case based reasoning and the learning automatamodel. In this new model a learning automata model is used as adaptation mechanism which adapts previouslyexperienced cases to the problem to be solved. The objective of the proposed model is to reduce the number ofiterations required to find a semi-optimum solution. The application is modeled as a set of independent tasks andthe heterogeneous computing system is modeled as a network of machines. Using computer simulation, it isshown that the combined model outperforms the model that only uses learning automata.

HIS’04 Abstracts: Session 3.6

Optimization

Comparison of Local Search Implementation Schemes in Hybrid Evolutionary Multiobjective OptimizationAlgorithmsHisao Ishibuchi and Kaname Narukawa

We compare two schemes of local search in hybrid evolutionary multiobjective optimization (EMO) algorithms. Oneis based on the weighted sum of multiple objectives, and the other is based on Pareto dominance. The twoschemes are compared with each other through computational experiments on a knapsack problem and ascheduling problem. We also show that a simple modification of the weighted sum-based local search schemeimproves the search ability of hybrid EMO algorithms.

A Two-Phase Genetic and Set Partitioning Approach for the Vehicle Routing Problem with Time WindowsGuilherme Bastos Alvarenga, Geraldo Robson Mateus

The Vehicle Routing Problem with Time Windows (VRPTW) is a well-known and complex combinatorial problem,which has received considerable attention in recent years. Results from exact methods have been improvedexploring parallel implementations and modern branch-and-cut techniques. However, 23 out of the 56 high orderinstances from Solomon’s test set still remain unsolved. Additionally, in many cases a prohibitive time is needed tofind the exact solution. Many efficient heuristic methods have been developed to make possible a good solution ina reasonable amount of time. Using travel distance as the main objective, this paper proposes a robust heuristicapproach for the VRPTW using an efficient genetic algorithm and a set partitioning formulation. The tests were runusing both, real numbers and truncated data type, making it possible to compare the results with previous heuristicand exact methods. Furthermore, computational results show that the proposed heuristic approach outperforms allprevious known heuristic methods in the literature, in terms of the minimal travel distance.

Hierarchical Tournament Selection Genetic Algorithm for the Vehicle Routing Problem with Time WindowsGuilherme Bastos Alvarenga, Geraldo Robson Mateus

The Vehicle Routing Problem with Time Windows (VRPTW) is a well-known and complex combinatorial problem,which has received considerable attention in recent years. The VRPTW benchmark problems of Solomon (1987)have been most commonly chosen to evaluate and compare all exact and heuristic algorithms. A genetic algorithmand a set partitioning two phases approach has obtained competitive results in terms of total travel distanceminimization. However, a great number of heuristics has used the number of vehicles as the first objective andtravel distance as the second, subject to the first. This paper proposes a three phases approach considering bothobjectives. Initially, a hierarchical tournament selection genetic algorithm is applied. It can reach all best results innumber of vehicles of the 56 Solomon’s problems explored in the literature. After then, the two phase approach,the genetic and the set partitioning, is applied to minimize the travel distance as the second objective.

How to Deal with the VRPTW by using Multi-Agent CoalitionsImen Boudali, Wajdi Fki, Khaled Ghedira

The Vehicle Routing Problem with Time Windows (VRPTW) is a well known combinatorial optimization problemoften met in many fields of industrial applications. We are interested in a coalition based multi-agent model (Coal-VRP) for the VRPTW. However, this model presents some drawbacks due to its spatial and temporal complexity.In order to overcome these drawbacks while maintaining the solution quality, we propose in this paper a newversion of this model called DyCoal-VRP. It is essentially based on dynamic generation of coalitions. Anexperimental validation of our model is achieved on the base of Solomon’s benchmark.

An Hybrid Approach for a Constrained Routing ProblemJesus Fabian Lopez

Regarding the solution of combinatorics problems, is important to evaluate the cost and the benefit amongobtaining solutions of high quality in detriment of the computational resources required. The problem presentedhere is about the routing of a vehicle with pickup and delivery of product with time window constraints. Thisproblem requires to be attended with instances of great scale (nodes >=100). We have a large quantity of nodeswith restrictions of time windows (>=90%) and with a large factor of amplitude (>=75) as well. The problem is NP-hard and for such motive the application of an exact method of solution to resolve it, is limited by the practical timefor routing. This paper proposes a specialized genetic algorithm, which offers solutions of good quality incomputational times that do useful its application.

HIS’04 Poster Abstracts

An Effective Machine Learning Algorithm using Momentum SchedulingEunmi Kim, Baeho Lee

This paper proposes learning performance improvement of support vector machine using the kernel relaxation andthe dynamic momentum. The dynamic momentum is reflected into different momentum according to the currentstate. While static momentum is equally influenced as a whole, the proposed dynamic momentum algorithm cancontrol the convergence rate and the performance according to the change of the dynamic momentum by training.The proposed algorithm has been applied to the kernel relaxation as the new sequential learning method ofsupport vector machine which has been presented recently. In order to show the efficiency of proposed algorithm,SONAR data, which is the neural network classifier standard evaluation data, is used. The simulation results ofproposed algorithm have better the convergence rate and performance than those using kernel relaxation andstatic momentum, respectively.

Design of Robust PID Controller With Disturbance Rejection For Motor Using Immune AlgorithmDong Hwa Kim, Jae Hoon Cho

In this paper, design approach of PID controller with rejection function against external disturbance in motor controlsystem is proposed using immune algorithm. Up to the present time, PID Controller has been used to operate forAC motor drive because of its implementational advantages in practice and simple structure. However, it is noteasy to achieve an optimal PID gain with no experience, since the gain of the PID controller has to be manuallytuned by trial and error in the industrial world. To design disturbance rejection function, disturbance rejectionconditions based on are illustrated and to decide the performance of response for the designed PID controller, anITSE (Integral of time weighted squared error) is used. Hence, parameters of PID controller are selected byimmune algorithm to obtain the required response.

Robust Motor Control Using clonal selection of Immune Algorithm Based MultiobjectiveDong Hwa Kim, Jae Hoon Cho, Hwan Lee

Strictly maintaining the steam temperature can be difficult due to heating value variation to the fuel source, timedelay changes in the main steam temperature, the change of the dynamic characteristics in the reheater. Up to thepresent time, PID Controller has been used to operate this system. However, it is very difficult to achieve anoptimal PID gain with no experience, since the gain of the PID controller has to be manually tuned by trial anderror. This paper focuses on tuning of the Controller with disturbance rejection using immune based multiobjectiveapproach. In this paper, an ITSE(Integral of time weighted squared error) is used to decide performance of tuningresults.

Realizing Small Involuntary Movements in Reading EnglishYukio Ishihara and Satoru Morita

The human eye continues to make extremely small movements even while looking at a single position. The eyes'behavior has been called small involuntary movements. A small involuntary movement occurs every 40ms andeach fixation lasts for 200ms in human eye movement. So, one fixation includes five small involuntary movements.The aim of this study is to realize fixations which include five small involuntary movements. The fixations arerealized on a computer by decreasing the capacity of the information that foveated vision obtains and by increasingthe frequency of short-distance eye movements. It is then shown that fixations which include more smallinvoluntary movements occur when the capacity is lower and the frequency is higher. Furthermore, we determinethe relationship between information capacity and movement frequency by using a computer to realize smallinvoluntary movements while reading English.

A Short Study on Dynamical Properties of Population based Incremental Incremental LearningR. Rastegar, M. R. Meybodi

In this paper, we study some dynamical properties of population-based incremental learning (PBIL) algorithm whenit uses truncation, proportional or Boltzmann selection schemas. Our analysis shows that if population size tends toinfinite, then with any learning rate, local optimum points of the function to be optimized are stable fixed points ofPBIL.

Fuzzy Reasoning in Sparse Fuzzy Rule BasesBaowen Wang, Xia Li, Wenyuan Liu and Yan Shi

In the sparse fuzzy rule bases, by Kóczy’s linear interpolative reasoning method, the inference consequences arenot always normal and convex fuzzy sets pointed by authors. For improving this problem, authors have proposedsome of applicable conditions. However, these conditions are still limited for real world applications of fuzzyreasoning system. Especially, as many times fuzzy inference in a sparse fuzzy rule base it needs many timescalculations, which means to increase the complexity of calculation and decrease the inference speed andefficiency of the fuzzy inference system. This paper presents a new fuzzy interpolative reasoning method in thesparse fuzzy rule bases based on so-called similarity relations of fuzzy sets. By this reasoning method, aninference consequence can be simply obtained, and is a normal and convex fuzzy set without any limitation, whichshows the potential ability of the proposed method in real-world fuzzy applications.

Stigmergy in Multi Agent Reinforcement LearningRaghav Aras, Alain Dutech, François Charpillet

In this paper, we review current uses of stigmergy in MARL and explore both analogies. Using stigmergy for multi-agent reinforcement learning (MARL), and applying existing MARL formulations for natural stigmergic phenomenasuch as ant piling or foraging. We comment on how certain aspects of stigmergy can be utilized in current MARLtechniques to speed up learning. We describe a memory-enabled MARL mechanism inspired by the phenomenaof stigmergy . One of the assumptions of MARL is that agents are memory-less. While inter-agent communicationfor coordination has received some attention in MARL, memory-based agents have not. Moreover, memory-basedreinforcement learning has been heretofore limited to single-agents using external memory to better operate inpartially-observable domains.

Field-Programmable Analog Filters Array with Applications for Fuzzy Inference SystemsSilviu Ionita, Emil Sofron

This paper approaches the issue of implementation of Fuzzy Inference System from the programmable analogcircuits’ perspective. Based on the actual paradigm of FPAA we consider an interesting and promising idea toimplement the membership functions using the frequency-response curves of the analog filters. The other fewfunctional blocks can be additionally implemented with analog and programmable analog structures in order toobtain a complete unit for fuzzy inference processing. Finally, we advance architecture suitable for implementationwith analog programmable structures.

A Waterfall Model for Knowledge Management and Experience ManagementZhaohao Sun

This paper examines experience and knowledge, experience management and knowledge management, and theirinterrelationships. It then proposes waterfall models for both experience management and knowledgemanagement. The models characterize knowledge management and experience management as the integration ofexperience processing and corresponding management, that of knowledge processing and correspondingmanagement respectively. The proposed approach will facilitate research and development of knowledgemanagement, experience management, and hybrid intelligent systems.

An Attitude Based Cooperative Negotiation ModelMadhu Goyal

In multi-agent setting agent teams often encounter conflicts in agents’ plans and actions. This paper presents acooperative negotiation model (ABCON) that allows agents in a team to appropriately negotiate various options ina hostile and dynamic fire world. It shows that negotiations explore the attitudes and behaviors that help agents tomanage conflict constructively. It says that cooperative negotiation is guided by the agents’ dynamic assessment ofalternative actions given the different scenario conditions.

Anatomy of Swarms for Clustering DataAlfred Ultsch

Systems for clustering with collectives of autonomous agents follow either the ant approach of picking up anddropping objects or the DataBot approach of identifying the data points with artificial life creatures. In DataBotsystems the clustering behaviour is controlled by movement programs. This paper reports the answers to twoquestions regarding movement: first, what is the most elementary movement program for clustering and second,what movement strategy can effectively solve even very difficult cluster problems. The clustering abilities aretested on synthetic data which are difficult to cluster. The effective movement strategy found is applied to realworld data of stock markets and DNA microarrays.

Investigation on the Threshold Setting of Intracellular Calcium Ion Concentration in Hippocampal CA3Network Model for Synaptic PlasticityKazuumi Tashiro, Noriaki Suetake and Eiji Uchino

It is reported that the histogram of an intracellular Ca2+ concentration ([Ca2+]i) is closely related to plasticity ofbiological cells. In this paper, in order to find a relation between a distribution of [Ca2+]i and an activity of CA3network model with synaptic plasticity, the distribution of [Ca2+]i is investigated. We propose in this paper apresumption for making the modification rule of interconnection plasticity among cells.

Reinforcement Learning of Player Agents in RoboCup Soccer SimulationAbhinav Sarje, Amit Chawre, Shivashankar B. Nair

In the past few years, Multi-Agent Systems (MAS) have emerged as an active subfield of Artificial Intelligence.Because of their inherent complexity, machine learning (ML) techniques have greatly aided the building of suchsystems. Robotic soccer has been used as a good domain for studying MAS and Machine Learning. In thischallenge, the task is to create learning and training methods to have the skills of a group of soccer playing agents.Just as young soccer players must learn to control the ball before learning any complex strategies, Robotic Socceragents must also acquire low-level skills before exhibiting complex behaviors: the more sophisticatedunderstanding of how to act as a part of the team becomes useless without the ability to execute the basicindividual tasks. In this paper, we train a `RoboCup' soccer agent, with the low level skill of `intercepting a movingball' using reinforcement learning. We describe the learning method in detail and report on our extensive simulationusing observations and results obtained.

Segmentation of High Resolution Satellite Images by Direction and Morphological FiltersTomoko Tateyama, Xian Yan Zeng, Zensho Nakao, Yen-Wei Chen

This paper examines images taken from IKONOS to extract several features such as road relations automatically.We propose a new method which combines color, texture information and shape information for segmentation ofhigh resolution satellite images. The method uses color and texture information for global segmentaiton, and shapeinformation for local analysis. We propose a new direction filter which pays its attention to road features havinginformation on specific directionality. We also propose another new morphology filter which is used as a lengthfilter extracting length of each region more efficiently.