mieczysław a. kłopotek, sławomir t. wierzchon, krzysztof
TRANSCRIPT
Mieczysław A. Kłopotek, Sławomir T. Wierzchon, Krzysztof Trojanowski (Eds.)
Intelligent Information Processing and Web Mining
Advances in Soft Computing
Editor-in-chiefProf. Janusz KacprzykSystems Research InstitutePolish Academy of Sciencesul. Newelska 601-447 WarsawPolandE-mail: [email protected]
Further volumes of this seriescan be found on our homepage:springer.com
Andrea Bonarini, Francesco Masulli andGabriella Pasi (Eds.)Soft Computing Applications, 2002ISBN 3-7908-1544-6
Leszek Rutkowski, Janusz Kacprzyk (Eds.)Neural Networks and Soft Computing, 2003ISBN 3-7908-0005-8
Jürgen Franke, Gholamreza Nakhaeizadeh,Ingrid Renz (Eds.)Text Mining, 2003ISBN 3-7908-0041-4
Tetsuzo Tanino, Tamaki Tanaka, MasahiroInuiguchiMulti-Objective Programming and GoalProgramming, 2003ISBN 3-540-00653-2
Mieczysław Kłopotek, Sławomir T.Wierzchon, Krzysztof Trojanowski (Eds.)Intelligent Information Processing and WebMining, 2003ISBN 3-540-00843-8
Ajith Abraham, Katrin Franke, MarioKöppen (Eds.)Intelligent Systems Design and Applications,2003ISBN 3-540-40426-0
Ahmad Lotfi, Jonathan M. Garibaldi (Eds.)Applications and Science in Soft-Computing,2004ISBN 3-540-40856-8
Mieczysław Kłopotek, Sławomir T.Wierzchon, Krzysztof Trojanowski (Eds.)Intelligent Information Processing and WebMining, 2004ISBN 3-540-21331-7
Miguel López-Díaz, Maríaç. Gil,Przemysław Grzegorzewski, OlgierdHryniewicz, Jonathan LawrySoft Methodology and Random InformationSystems, 2004ISBN 3-540-22264-2
Kwang H. LeeFirst Course on Fuzzy Theory andApplications, 2005ISBN 3-540-22988-4
Barbara Dunin-Keplicz, Andrzej Jankowski,Andrzej Skowron, Marcin SzczukaMonitoring, Security, and RescueTechniques in Multiagent Systems, 2005ISBN 3-540-23245-1
Bernd Reusch (Ed.)Computational Intelligence, Theory andApplications: International Conference 8thFuzzy Days in Dortmund, Germany,Sept. 29 – Oct. 01, 2004 Proceedings, 2005ISBN 3-540-2280-1
Frank Hoffmann, Mario Köppen, FrankKlawonn, Rajkumar Roy (Eds.)Soft Computing: Methodologies andApplications, 2005ISBN 3-540-25726-8
Ajith Abraham, Bernard de Baets, MarioKöppen, Bertram Nickolay (Eds.)Applied Soft Computing Technologies: TheChallenge of Complexity, 2006ISBN 3-540-31649-3
Ashutosh Tiwari, Joshua Knowles, ErelAvineri, Keshav Dahal, Rajkumar Roy (Eds.)Applications of Soft Computing, 2006ISBN 3-540-29123-7
Mieczysław A. Kłopotek, Sławomir T.Wierzchon, Krzysztof Trojanowski (Eds.)Intelligent Information Processing and WebMining, 2006ISBN 3-540-33520-X
Mieczysław A. KłopotekSławomir T. WierzchonKrzysztof Trojanowski(Eds.)
Intelligent InformationProcessing and Web MiningProceedings of the International IIS: IIPWM’06Conference held in Ustron, Poland, June 19-22, 2006
ABC
Mieczysław A. KłopotekSławomir T. WierzchonKrzysztof TrojanowskiPolish Academy of SciencesInstitute of Computer Scienceul. Ordona 21, 01-237Warszawa, PolandE-mail: [email protected]
Library of Congress Control Number: 2006923821
ISSN print edition: 1615-3871ISSN electronic edition: 1860-0794ISBN-10 3-540-33520-X Springer Berlin Heidelberg New YorkISBN-13 978-3-540-33520-7 Springer Berlin Heidelberg New York
This work is subject to copyright. All rights are reserved, whether the whole or part of the material isconcerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publicationor parts thereof is permitted only under the provisions of the German Copyright Law of September 9,1965, in its current version, and permission for use must always be obtained from Springer. Violations areliable for prosecution under the German Copyright Law.
Springer is a part of Springer Science+Business Mediaspringer.comc© Springer-Verlag Berlin Heidelberg 2006
Printed in The Netherlands
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,even in the absence of a specific statement, that such names are exempt from the relevant protective lawsand regulations and therefore free for general use.
Typesetting: by the authors and techbooks using a Springer LATEX macro packageCover design: Erich Kirchner, Heidelberg
Printed on acid-free paper SPIN: 11737841 89/techbooks 5 4 3 2 1 0
Preface
This volume contains selected papers, presented at the international con-
ference on Intelligent Information Processing and Web Mining Conference
IIS:IIPWM’06, organized in Ustroń (Poland) on June 19-22nd, 2006.
The event was organized by the Institute of Computer Science of Polish
Academy of Sciences, a leading Polish research institution in fundamental
and applied research in the area of Artificial Intelligence (AI) and of Infor-
mation Systems (IS), in cooperation with a number of scientific and business
institutions. It was a continuation of a series of conferences on these subjects,
initiated by Prof. M. Dąbrowski and Dr. M. Michalewicz in 1992.
The conference was addressed primarily to those who are active in Ar-
tificial Immune Systems (AIS) and other biologically motivated methods,
Computational Linguistics (CL), Web technologies (WT), and Knowledge
Discovery (KD), and all kinds of interactions between those fields.
The submitted papers covered new computing paradigms, among others in
biologically motivated methods, advanced data analysis, new machine learn-
ing paradigms, natural language processing, new optimization technologies,
applied data mining using statistical and non-standard approaches. The pa-
pers give an overview over a wide range of applications for intelligent systems:
in operating systems design, in network security, for information extraction
from multimedia (sound, graphics), in financial market analysis, in medicine,
in geo-science, etc.
Though numerous papers have been submitted, only a fraction of them
(about 40%) was accepted for publication in a rigorous reviewing process.
At this point we would like to express our thanks to the members of
Programme Committee, as well as additional reviewers, who kindly agreed
to express their opinions, for their excellent job. Also we are thankful to the
organizers of the special sessions accompanying this conference. But first of
all we are deeply indebted the contributors to this volume and those whose
papers were not qualified for publication for their hard research work that
made this conference such an exciting event.
On behalf of the Program Committee and of the Organizing Committee
we would like to thank all participants: computer scientists, mathematicians,
engineers, logicians and other interested researchers who found excitement
in advancing the area of intelligent systems. We hope that this volume of
VI
IIS:IIPWM’06 Proceeding will be a valuable reference work in your further
research.
We would like to thank Dr. M. Wolinski for his immense effort in resolving
technical issues connected with the preparation of this volume.
Ustroń, Poland Mieczysław A. Kłopotek, Conference Co-Chair
June 2006 Sławomir T. Wierzchoń, Conference Co-Chair
Krzysztof Trojanowski, Organizing Committee Chair
Preface
VII
We would like to thank to the Programme Committee Members for their
great job of evaluating the submissions:
• Witold Abramowicz (Poznan University of Economics, Poland)
• David Bell (Queen’s University Belfast, UK)
• Peter J. Bentley (University College London, UK)
• Petr Berka (University of Economics Prague, Czech Republic)
• Leonard Bolc (Polish Academy of Science, Poland)
• Damir Cavar (University of Zadar, Croatia)
• Vincenzo Cutello (University of Catania, Italy)
• Andrzej Czyzewski (Gdansk University of Technology, Poland)
• Piotr Dembiński (Polish Academy of Sciences, Poland)
• Włodzisław Duch (Nicholas Copernicus University, Poland)
• Nelson F. F. Ebecken (COPPE/Federal University of Rio de Janeiro,
Brazil)
• Tapio Elomaa (Tampere University of Technology, Finland)
• Floriana Esposito (Bary University, Italy)
• Jerzy W. Grzymała-Busse (University of Kansas, USA)
• Mohand-Saïd Hacid (Université Claude Bernard Lyon 1, France)
• Ray J. Hickey (University of Ulster, UK)
• Erhard Hinrichs (University of Tuebingen, Germany)
• Tu Bao Ho (Japan Advanced Institute of Science and Technology, Japan)
• Olgierd Hryniewicz (Polish Academy of Sciences, Poland)
• Janusz Kacprzyk (Polish Academy of Sciences, Poland)
• Samuel Kaski (Helsinki University of Technology, Finland)
• Jan Komorowski (Uppsala University, Sweden)
• Józef Korbicz (University of Zielona Góra, Poland)
• Jacek Koronacki (Polish Academy of Sciences, Poland)
• Bożena Kostek (Gdansk University of Technology, Poland)
• Geert-Jan M. Kruijff (German Research Center for Artificial Intelligence
(DFKI), Germany)
• Stan Matwin (University of Ottawa, Canada)
• Ernestina Menasalvas (Technical University of Madrid, Spain)
• Detmar Meurers (Ohio State University, USA)
• Maciej Michalewicz (NuTech Solutions Polska, Poland)
• Zbigniew Michalewicz (University of Adelaide, Australia)
• Ryszard S. Michalski (George Mason University, USA)
• Giuseppe Nicosia (University of Catania, Italy)
• Zdzisław Pawlak (Scientific Research Committee, Poland)
• James F. Peters (University of Manitoba, Canada)
• Adam Przepiórkowski (Polish Academy of Sciences, Poland)
• Zbigniew W. Raś (University of North Carolina at Charlotte, USA)
• Jan Rauch (University of Economics, Czech Republic)
• Gilbert Ritschard (University of Geneva, Switzerland)
• Henryk Rybiński (Warsaw University of Technology, Poland)
Programme Committee Members
VIII
• Abdel-Badeeh M. Salem (Ain Shams University, Egypt)
• Kiril Simov (Bulgarian Academy of Science, Bulgaria)
• Andrzej Skowron (Warsaw University, Poland)
• Tomek Strzałkowski (University At Albany, USA)
• Roman Świniarski (San Diego State University, USA)
• Stan Szpakowicz (University of Ottawa, Canada)
• Ryszard Tadeusiewicz (University of Science and Technology, Poland)
• Jonathan Timmis (University of York, UK)
• Zygmunt Vetulani (Adam Mickiewicz University, Poland)
• Alicja Wakulicz-Deja (University of Silesia, Poland)
• Hui Wang (University of Ulster, UK)
• Jan Węglarz (Poznan University of Technology, Poland)
• Stefan Węgrzyn (Polish Academy of Sciences, Poland)
• Alessandro Zanasi (TEMIS, Italy)
• Zhi-Hua Zhou (Nanjing University, China)
• Krzysztof Zieliński (University of Science and Technology, Poland)
• Djamel A. Zighed (Lumière Lyon 2 University, France)
We would like also to thank additional reviewers:
• Stanisław Ambroszkiewicz (Polish Academy of Sciences, Poland)
• Małgorzata Marciniak (Polish Academy of Sciences, Poland)
• Agnieszka Mykowiecka (Polish Academy of Sciences, Poland)
We are also indebted to the invited speaker:
• Walt Truszkowski (NASA Goddard Space Flight Center, USA)
and to organizers of invited sessions:
• Vincenzo Cutello (University of Catania, Italy)
• Giuseppe Nicosia (University of Catania, Italy)
• Adam Przepiórkowski (Polish Academy of Sciences, Poland)
• Henryk Rybiński (Warsaw University of Technology, Poland)
• Alicja Wakulicz-Deja (University of Silesia, Poland)
for their important contribution to the success of the conference.
Programme Committee Members
Part I. Regular Sessions: Artificial Immune Systems
Comparing Energetic and Immunological Selection in Agent-3
Aleksander Byrski, Marek Kisiel-Dorohinicki
11
Krzysztof Cetnarowicz, Renata Cięciwa, Gabriel Rojek
21
Controlling Spam: Immunity-based Approach . . . . . . . . . . . . . . . . . 31
Konrad Kawecki, Franciszek Seredyński, Marek Pilski
A Comparison of Clonal Selection Based Algorithms for Non-Stationary Optimisation Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Krzysztof Trojanowski, Sławomir T. Wierzchoń
Part II. Regular Sessions: Evolutionary Methods
On Asymptotic Behaviour of a Simple Genetic Algorithm . . . . 55
Witold Kosiński, Stefan Kotowski, Jolanta Socała
Evolutionary Algorithm of Radial Basis Function Neural Net-works and Its Application in Face Recognition . . . . . . . . . . . . . . . . 65
Jianyu Li, Xianglin Huang, Rui Li, Shuzhong Yang, Yingjian Qi
GAVis System Supporting Visualization, Analysis and Solv-ing Combinatorial Optimization Problems Using EvolutionaryAlgorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Piotr Świtalski, Franciszek Seredyński, Przemysław Hertel
Part III. Regular Sessions: Computational Linguistics
Gazetteer Compression TechniqueBased on Substructure Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Jan Daciuk, Jakub Piskorski
Table of Contents
to Secutrity Mechanisms in a Multiagent System . . . . . . . . . . . . .
Randomized Dynamic Generation of Selected Melanocytic Skin
Zdzisław S. Hippe, Jerzy W. Grzymała-Busse, Ł. PiątekLesion Features
Based Evolutionary Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
An Immunological and an Ethically-social Approach
X
WFT – Context-Sensitive Speech Signal Representation . . . . . . 97
Jakub Gałka, Michał Kępiński
Part IV. Regular Sessions: Web Technologies
Krzysztof Ciesielski, Michał Dramiński, Mieczysław A. Kłopotek,Dariusz Czerski, Sławomir T. Wierzchoń
Faster Frequent Pattern Mining from the Semantic Web . . . . . . 121
Joanna Józefowska, Agnieszka Ławrynowicz, Tomasz Łukaszewski
Collective Behaviour of Cellular Automata
Miroslaw Szaban, Franciszek Seredyński, Pascal Bouvry
Part V. Regular Sessions: Foundations of Knowledge Discovery
Experiments on Data with Three Interpretations of MissingAttribute Values—A Rough Set Approach . . . . . . . . . . . . . . . . . . . . 143
Jerzy W. Grzymała-Busse, Steven Santoso
Tableaux Method with Free Variables for Intuitionistic Logic . 153
Boris Konev, Alexander Lyaletski
A Similarity Measure between Tandem Duplication Trees . . . . 163
Jakub Koperwas, Krzysztof Walczak
Finding Optimal Decision Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Petr Máša, Tomáš Kočka
Attribute Number Reduction Process and Nearest NeighborMethods in Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Aleksander Sokołowski, Anna Gładysz
The Use of Compound Attributes in AQ Learning . . . . . . . . . . . . 189
Janusz Wojtusiak, Ryszard S. Michalski
Part VI. Regular Sessions: Statistical Methods in KnowledgeDiscovery
Residuals for Two-Way Contingency Tables, Especially ThoseComputed for Multiresponces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Guillermo Bali Ch., Dariusz Czerski, Mieczysław A. Kłopotek,Andrzej Matuszewski
Table of Contents
Adaptive Document Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131and Symmetric Key Cryptography.Rules
XI
Wieslaw Szczesny, Marek Wiech
Bartłomiej Śnieżyński
Part VII. Regular Sessions: Knowledge Discovery in Applications
Improving Quality of Agglomerative Scheduling
Pawel Boinski, Konrad Jozwiak, Marek Wojciechowski,Maciej Zakrzewicz
Analysis of the Structure of Online Marketplace Graph . . . . . . . 243
Andrzej Dominik, Jacek Wojciechowski
Trademark Retrieval in the Presence of Occlusion . . . . . . . . . . . . 253
Dariusz Frejlichowski
On Allocating Limited Sampling Resources Using a LearningAutomata-based Solution to the Fractional Knapsack Problem 263
Ole-Christoffer Granmo, B. John Oommen
Learning Symbolic User Models for Intrusion Detection: AMethod and Initial Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Ryszard S. Michalski, Kenneth A. Kaufman, Jaroslaw Pietrzykowski,Bartłomiej Śnieżyński, Janusz Wojtusiak
Kazuyoshi Miyara, Thai Duy Hien, Hanane Harrak, Yasunori Nagata,Zensho Nakao
Developing a Model Agent-based Airline Ticket AuctioningSystem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
Mladenka Vukmirovic, Maria Ganzha, Marcin Paprzycki
Multi-Label Classification of Emotions in Music . . . . . . . . . . . . . . 307
Alicja Wieczorkowska, Piotr Synak, Zbigniew W. Raś
Wrapper Maintenance for Web-Data Extraction Based
Shunxian Zhou, Yaping Lin, Jingpu Wang, Xiaolin Yang
Table of Contents
. . . . . . . . . . . 211Cluster Analysis and Generalized Association PlotsVisualizing Latent Structures in Grade Correspondence
Converting a Naive Bayes Model into a Set of Rules . . . . . . . . . . 221
in Concurrent Processing of Frequent Itemset Queries . . . . . . . . 233
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287EigenimagesMultichannel Color Image Watermarking Using PCA
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317on Pages Features
XII
Part VIII. Poster Session
Parsing Polish as a Context-Free Language . . . . . . . . . . . . . . . . . . . 329
Stanisław Galusa
María Adela Grando, Christopher D. Walton
Definiteness of Polish Noun Phrases Modified by Relative
Elżbieta Hajnicz
Hanane Harrak, Thai Duy Hien, Yasunori Nagata, Zensho Nakao
Tomasz Obrębski
Network Traffic Analysis Using Immunological
Marek Ostaszewski, Franciszek Seredyński, Pascal Bouvry
Event Detection in Financial Time Series
Tomasz Pelech, Jan T. Duda
Automation of Communication and Co-operation Processes in
Zbigniew Pietrzykowski, Jaroslaw Chomski, Janusz Magaj,Grzegorz Niemczyk
Predictive Analysis of the pO2 Blood Gasometry Parameter
Wieslaw Wajs, Mariusz Swiecicki, Piotr Wais, Hubert Wojtowicz,Pawel Janik, Leszek Nowak
Application of Fuzzy Logic Theory to Geoid Heightnation
Mehmet Yılmaz, Mustafa Acar, Tevfik Ayan, Ersoy Arslan
Part IX. Invited Session: Knowledge Base Systems
On Greedy Algorithms with Weights for Constructionof ParMikhail Ju. Moshkov, Marcin Piliszczuk, Beata Zielosko
Table of Contents
MAP : a Language for Modelling Conversations in AgentEnvironments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Clauses in the DRT Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
DCT Watermarking Optimization by Genetic Programming . . 347
Searching Text Corpora with grep . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
by Immune-Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Related to the Infants Respiration Insufficiency . . . . . . . . . . . . . . . 377
and Evolutionary Paradigms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371in Marine Navigation
Determi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383. .
tial Covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391. .
XIII
Barbara Marszał-Paszek, Piotr Paszek
Agnieszka Nowak, Alicja Wakulicz-Deja
Roman Siminski
Agnieszka Nowak, Roman Siminski, Alicja Wakulicz-Deja
Wojciech Froelich
Magdalena Alicja Tkacz
Part X. Invited Session: Applications of Artificial Immune Systems
Thomas Stibor, Jonathan Timmis, Claudia Eckert
How Can We Simulate Something As Complex As the Im-
Simon Garrett, Martin Robbins
Vincenzo Cutello, Giuseppe Nicosia, Emilio Pavia
Part XI. Invited Session: Data Mining – Algorithmsand Applications
Data Mining Approach to Classification of Archaeological Aerial
Łukasz Kobyliński, Krzysztof Walczak
Marzena Kryszkiewicz, Łukasz Skonieczny
Robert Bembenik, Henryk Rybiński
Table of Contents
Minimal Templates Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
The Inference Processes on Clustered Rules . . . . . . . . . . . . . . . . . . 403
Extending Decision Units Conception Using Petri Nets . . . . . . . 413
Towards Modular Representation of Knowledge Base . . . . . . . . . 421
Lazy Learning of Agent in Dynamic Environment . . . . . . . . . . . . . 429
Artificial Neural Network Resistance to Incomplete Data . . . . . 437
Generalization Regions in Hamming Negative Selection . . . . . . . 447
mune System? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
A Parallel Immune Algorithm for Global Optimization . . . . . . . 467
Photographs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
Hierarchical Document Clustering Using Frequent Closed Sets 489
Mining Spatial Association Rules with no Distance Parameter 499
XIV
Morfeusz — a Practical Tool for the Morphological Analysis
Marcin Woliński
Domain–Driven Automatic Spelling Correction
Agnieszka Mykowiecka, Małgorzata Marciniak
Maciej Piasecki, Grzegorz Godlewski
Index of Authors
Table of Contents
of Polish . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
Reductionistic, Tree and Rule Based Tagger for Polish . . . . . . . . 531
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
and Morphosyntactic Processing of PolishPart XII. Invited Session: Fundamental Tools for the Lexical
for Mammography Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
Part I
Regular Sessions: Artificial Immune Systems
Comparing Energetic and ImmunologicalSelection in Agent-Based Evolutionary
Optimization
Aleksander Byrski and Marek Kisiel-Dorohinicki
Department of Computer ScienceAGH University of Science and Technology, Krakow, Poland{olekb,doroh}@agh.edu.pl
Abstract. In the paper the idea of an immunological selection mechanism forthe agent-based evolutionary computation is presented. General considerations areillustrated by the particular system dedicated to function optimization. Selectedexperimental results allow for the comparison of the performance of immune-inpiredselection mechanisms and classical energetic ones.
1 Introduction
The idea of agent-based evolutionary optimization most generally consists inthe incorporation of evolutionary processes into a multi-agent system at apopulation level. In its fine-grained model it means that besides interactionmechanisms typical for agent-based systems (such as communication) agentsare able to reproduce (generate new agents) and may die (be eliminatedfrom the system). Inheritance is accomplished by an appropriate definition ofreproduction (with mutation and recombination), which is similar to classicalevolutionary algorithms. Selection mechanisms correspond to their naturalprototype and are based on the existence of non-renewable resource calledlife energy, which is gained and lost when agents perform actions [5].
These so called evolutionary multi-agent systems (EMAS) proved workingin a number of applications. Yet still they reveal new features, particularlywhen supported by specific mechanisms borrowed from other methods knownin the soft computing area [6]. Following this idea, immunological approachwas proposed as a more effective alternative to the classical energetic selectionused in EMAS [1]. Introduction of immune-based selection mechanisms mayaffect several aspects of the system behaviour, such as the diversity of thepopulation and the dynamics of the whole process. This paper focuses on theimpact of the immune-based approach on the performance of EMAS appliedto function optimization in comparison to the classical energetic selectionused alone.
Below, after a short presentation of the basics of human immunity andartificial immune systems, the details of the proposed approach are given.Then comes the discussion of the results obtained, which allow for someconclusions to be drawn.
A. Byrski and M. Kisiel-Dorohinicki: Comparing Energetic and Immunological Selection in
www.springerlink.com c© Springer-Verlag Berlin Heidelberg 2006Agent-Based Evolutionary Optimization, Advances in Soft Computing 5, 3–10 (2006)
4 Aleksander Byrski and Marek Kisiel-Dorohinicki
NEGATIVE SELECTION CLASSIFICATION
ImmatureT-cells
match
no match
MatureT-cells
match
no match
NON-SELF
SELF
SELFPROTEINS UNKNOWN PROTEINS
Fig. 1. Negative selection mechanism in artificial immune systems
2
Human immune system plays a key role in maintaining the stable functioningof the body. It allows for detection and elimination of disfunctional endoge-nous cells, termed infectious cells and exogenous microorganisms, infectiousnon-self cells such as bacteria and viruses, which enter the body through vari-ous routes, including the respiratory, digestive systems, and damaged dermaltissues. A key role in humoral immunity (a cellular immune layer) play lym-phocytes. T-lymphocytes mature in thymus into two distinct subpopulations:T-helper and T-killer cells, the latter acting as removing agents for disfunc-tional cells of the body. T-cells are subjected to a process called negativeselection in thymus, where they are exposed to a wide variety of self proteins,and destroyed if they recognize them [4].
Artificial immune systems, inspired by the human immunity, recently be-gan to be the subject of increased researchers’ interest. Different immune-inspired approaches were applied to many problems, such as classification oroptimization [9]. The most often used algorithm of negative selection corre-sponds to its origin and consists of the following steps (see fig. 1):
1. Lymphocytes are created, as yet they are considered immature.2. The binding of these cells (affinity) to present self-cells (eg. good solutions
of some problem) is evaluated.3. Lymphocytes that bind themselves to good” cells are eliminated.4. Lymphocytes that survive are considered mature.
Mature lymphocytes are presented with the cells that have unknown origin(they may be self, or non-self cells), and they are believed to have possibilityof classifying them [8].
Artificial Immune Systems
“
Comparing Energetic and Immunological Selection 5
Immune-based algorithms may be used also in optimization problems –one of such approaches is known as the Artificial Immune Iterated Algorithm(AIIA) and was originally presented in [3] and modified in [7]. The algorithmconsists of following steps:
1. The population of individuals (called antibodies) is randomly generated.Each individual represents a single solution in the search space.
2. The best antibody is chosen (antigen).3. A group of antibodies is selected with the highest affinity (similarity) to
the antigen (clonal selection).4. Each individual is cloned and mutated, if the best clone is better than
the original – the original is replaced (somatic hyper mutation).5. Individuals with low fitness are replaced by randomly generated new ones
(apoptosis).
In this way good solutions of the problem are retained in the population, andthe whole population is attracted by the currently chosen antigen.
3
Selection mechanisms known from classical evolutionary computation cannotbe used in evolutionary multi-agent systems because of the assumed lack ofglobal knowledge (which makes it impossible to evaluate all individuals atthe same time), and the autonomy of agents (which causes that reproductionis achieved asynchronously). The resource-based (energetic) selection schemeassumes that agents are rewarded for good” behaviour, and penalized for”bad” behaviour (which behaviour is considered good” or bad” depends onthe particular problem to be solved) [5]. In the simplest case the evaluationof an agent (its phenotype) is based on the idea of agent rendezvous. Assum-ing some neighbourhood structure in the environment, agents evaluate theirneighbours, and exchange energy. Worse agents (considering their fitness) areforced to give a fixed amount of their energy to their better neighbours. Thisflow of energy causes that in successive generations, survived agents shouldrepresent better approximations of the solution [2].
In order to speed up the process of selection, based on the assumptionthat bad” phenotypes come from the bad” genotypes, a new group of agents(acting as lymphocyte T-cells) may be introduced [1]. They are responsible forrecognizing and removing agents with genotypes similar to the genotype pat-tern posessed by these lymphocytes. Other approach may introduce specificpenalty applied by T-cells for recognized agents (certain amount of agent’s en-ergy is removed) instead of removing them from the system. Of course theremust exist some predefined affinity function, which may be based on thepercentage difference between corresponding genes. The agents-lymphocytesmay be created in the system in two ways:
From Evolutionary to Immunological Multi-agentSystems
““ “
“ “
6 Aleksander Byrski and Marek Kisiel-Dorohinicki
ImmuneManager
Agent
Lymphocyte
Agent
Agent
evaluation
tra
nsfo
rma
tio
n
testing
removingre
mo
vin
g
communication
Fig. 2. Immunological selection principle in iEMAS
1. vaccination—during system initialisation lymphocytes are created withrandom genotype patterns, or with the patterns generated by some othertechnique designed to solve a similar problem,
formed into lymphocyte patterns by means of mutation operator, andthe newly created group of lymphocytes is introduced into the system.
In both cases, new lymphocytes must undergo the process of negative selec-tion. In a specific period of time, the affinity of immature lymphocytes pat-terns to good” agents (posessing relative high amount of energy) is tested.If it is high (lymphocytes recognize good” agents as non-self”) they are re-moved from the system. If the affinity is low, it is assumed that they will beable to recognize
”
non-self” individuals (
”
bad” agents) leaving agents withhigh energy intact.
4
The system was designed and implemented using distributed evolutionarymulti-agent platform AgE developed at AGH-UST1.
In the proposed approach, real-valued encoding of the solutions is used,and the affinity is determined in the following way. Let p = [p1 . . . pn] be aparatope (genotype pattern owned by the lymphocyte) and e = [e1 . . . en] bean epitope (genotype owned by the antigen—in the described system it is
1 http://age.iisg.agh.edu.pl
““ “
Implementation of Immunological EMASfor
2. causality—after the action of death, the late agent genotype is trans-
Optimization
Comparing Energetic and Immunological Selection 7
simply the genotype of the tested agent), and n is the length of the geno-type. If dmax is the maximum deviation and cmin is the minimum count ofcorresponding genes, the construction of the set of corresponding genes maybe considered:
G = {pi :∣∣∣∣pi
ei
∣∣∣∣ ≤ dmax}
The lymphocyte is considered as stimulated (its affinity reached minimallevel) when
=
G ≥ cmin, and considered as non-stimulated otherwise.
5
The experiments were performed in order to show whether the introductionof immune-based mechanisms into EMAS will affect the classical energeticselection. Five-dimensional Rastrigin and Rosenbrock functions were used asbenchmark optimization problems.
The system consisted of five evolutionary islands, with initial populationof 30 agents on every island. Immune T-cells affected energy of the agents in-stead of removing them from the environment—after successful affinity bind-ing, certain amount of agent’s energy was transferred to the system (dividedamong other agents).
The plots discussed below show averaged values (with the estimates ofstandard deviation) obtained for 20 runs of the system with the same param-eters.
In figures 3a and 4a, averaged best fitness is presented in consecutive stepsof the system activity for EMAS and iEMAS. It seems that introduction ofthe immunological selection mechanism does not affect the quality of obtainedresults. Both systems (EMAS and iEMAS) reached sub-optimal values in theobserved period of time (continuation of the search would yield better results,however it was not the primary task of this research). Yet looking at figures 3band 4b, which present the number of agents in consecutive steps of the systemactivity, it may be observed that the introduction of T-cells affect greatly thedynamics of the population. The number of agents in iEMAS is noticablysmaller than in EMAS, which means that a similar result is achieved usingsmaller population—the effectiveness of the search increases.
In order to measure the diversity of the population, a specific coefficientwas proposed:
Xi = [xi1, . . . , xiM ], i ∈ [1 . . .N ],
αj =
√∑Ni=1(xij − xi)2
N − 1
σ =
√∑Mj=1(αj − αj)2
M − 1
Experimental Results
8 Aleksander Byrski and Marek Kisiel-Dorohinicki
5
10
15
20
25
30
35
40
45
0 500 1000 1500 2000 2500 3000
Fitn
ess
Step
a) EMAS
5
10
15
20
25
30
35
40
45
0 500 1000 1500 2000 2500 3000
Fitn
ess
Step
a) EMASiEMAS
0
100
200
300
400
500
0 500 1000 1500 2000 2500 3000
Age
nt c
ount
Step
b) EMAS
0
100
200
300
400
500
0 500 1000 1500 2000 2500 3000
Age
nt c
ount
Step
b) EMASiEMAS
0.1
1
10
100
0 500 1000 1500 2000 2500 3000
Div
ersi
ty
Step
c) EMAS
0.1
1
10
100
0 500 1000 1500 2000 2500 3000
Div
ersi
ty
Step
c) EMASiEMAS
0
10
20
30
40
50
60
0 500 1000 1500 2000 2500 3000
TC
ell c
ount
Step
d) iEMAS
Fig. 3. The best fitness (a), agent count (b), diversity (c) and T-Cell count (d) insystem steps for Rastrigin function optimization
where Xi is the genotype of the certain individual. The σ coefficient maybe perceived as the estimate of the standard deviation of the estimate ofstandard deviation of the average vector coordinates (αj). Looking at figures3c and 4c, one may notice that the diversity was preserved in both cases.In the beginning of search, a specific raise of diversity may be observed forEMAS system, which is caused by the randomness of the initial populationof individuals. This effect is suppressed in iEMAS by the introduction ofT-Cells.
Figures 3d and 4d show the number of T-Cells introduced into the systemin consecutive steps of its activity. It may be seen that after initial significantoscillations (caused by dynamic changes of the population size in the begin-ning of the computation) the number of T-Cells falls down and stabilizes, sothe system is not dominated by these agents (which would affect the overallefficiency of the computation).
6 Conclusion
In the paper immune-based selection mechanisms for evolutionary multi-agent systems were evaluated to show their performance in comparison toclassical energetic ones. As the experimental results show, it lowers the cost
Comparing Energetic and Immunological Selection 9
5
10
15
20
25
30
35
40
45
0 500 1000 1500 2000 2500 3000
Fitn
ess
Step
a) EMAS
5
10
15
20
25
30
35
40
45
0 500 1000 1500 2000 2500 3000
Fitn
ess
Step
a) EMASiEMAS
0
100
200
300
400
500
0 500 1000 1500 2000 2500 3000
Age
nt c
ount
Step
b) EMAS
0
100
200
300
400
500
0 500 1000 1500 2000 2500 3000
Age
nt c
ount
Step
b) EMASiEMAS
0.1
1
10
100
0 500 1000 1500 2000 2500 3000
Div
ersi
ty
Step
c) EMAS
0.1
1
10
100
0 500 1000 1500 2000 2500 3000
Div
ersi
ty
Step
c) EMASiEMAS
0
10
20
30
40
50
60
0 500 1000 1500 2000 2500 3000
TC
ell c
ount
Step
d) iEMAS
Fig. 4. The best fitness (a), agent count (b), diversity (c) and T-Cell count (d) insystem steps for Rosenbrock function optimization
of the computation by removing useless solutions (the population of agentsis smaller), though the results are comparable to these obtained for the sys-tem without immunological selection. Additionally it was shown, that theimmune-based approach still preserves the diversity of the population at sim-ilar level as for the classical energetic selection, the T-Cells population doesnot dominate the system, which would affect its performance.
Further research should allow to compare the discussed method with clas-sical methods of immune-based and evolutionary optimization. Especially thebehaviour of the system for difficult multi-dimensional problems will be ver-ified. Also the influence of the particular parameters on the performance ofthe search will be evaluated.
References
1. Aleksander Byrski and Marek Kisiel-Dorohinicki. Immunological selection mech-anism in agent-based evolutionary computation. In M. K�lopotek, S. Wierzchon,and K. Trojanowski, editors, Intelligent Information Processing and Web Min-ing, Advances in Soft Computing, pages 411–415. Springer-Verlag, 2005.
2. Grzegorz Dobrowolski and Marek Kisiel-Dorohinicki. Management of evolution-ary MAS for multiobjective optimization. In Tadeusz Burczynski and Andrzej
10 Aleksander Byrski and Marek Kisiel-Dorohinicki
Osyczka, editors, Evolutionary Methods in Mechanics, pages 81–90. Kluwer Aca-demic Publishers, 2004.
3. A. Gaspar and Ph. Collard. From GAs to artificial immune systems: Improvingadaptation in time dependent optimisation. In Proc. of the 1999 Congress onEvolutionary Computation – CEC’99. IEEE Publishing, 1999.
4. W.H. Johnson, L.E. DeLanney, and T.A. Cole. Essentials of Biology. New York,Holt, Rinehart and Winston, 1969.
5. Marek Kisiel-Dorohinicki. Agent-oriented model of simulated evolution. InWilliam I. Grosky and Frantisek Plasil, editors, SofSem 2002: Theory and Prac-tice of Informatics, volume 2540 of Lecture Notes in Computer Science. Springer-Verlag, 2002.
6. Marek Kisiel-Dorohinicki, Grzegorz Dobrowolski, and Edward Nawarecki. Agentpopulations as computational intelligence. In Leszek Rutkowski and JanuszKacprzyk, editors, Neural Networks and Soft Computing, Advances in Soft Com-puting, pages 608–613. Physica-Verlag, 2003.
7. K. Trojanowski and S. Wierzchon. Studying properties of multipopulationheuristic approach to non-stationary optimisation tasks. In Proc. of the Int.Conf. Intelligent Information System, Intelligent Information Processing andWeb Mining. Springer Verlag, 2003.
8. S. Wierzchon. Deriving concise description of non-self patterns in an artificialimmune system. in: L.C. Jain, J. Kacprzyk, editors, New Learning Paradigm inSoft Comptuning. Physica-Verlag 2001.
9. S. Wierzchon. Function optimization by the immune metaphor. Task Quaterly,6(3):1–16, 2002.
An Immunological and an Ethically-socialApproach to Security Mechanisms in aMultiagent System
Krzysztof Cetnarowicz1, Renata Cięciwa2, and Gabriel Rojek3
1 Institute of Computer Science
AGH University of Science and Technology
Al. Mickiewicza 30, 30-059 Kraków, Poland
[email protected] Department of Computer Networks
Nowy Sącz School of Business — National-Louis University
ul. Zielona 27, 33-300 Nowy Sącz, Poland
[email protected] Department of Computer Science in Industry
AGH University of Science and Technology
Al. Mickiewicza 30, 30-059 Kraków, Poland
Abstract. This article presents a discussion about security mechanisms in agent
and multiagent systems. Presented discussion focuses on the design of an artificial
immune system for intrusion detection in agent systems. An immunological ap-
proach to change detection seems very useful in design of security mechanisms for
an agent functioning in his environment. Reasons for this expectation are the prin-
ciples of a computer immune system such as distribution and autonomy. Mentioned
principles of artificial immune systems are strongly connected with main principles
of agent technology which are the autonomy of an agent and distribution in the
case of multiagent system.
1 Agents and Multiagent Systems
Considering an immunological approach to intrusion detection in agent sys-
tems it has to be defined what exactly an agent is, but in the scientific sphere
of intelligent agents there is no agreement in this matter. The reason of this
controversy is that various attributes associated with agency are of different
meaning for different domains. One of most cited definitions of an agent is
presented in [15] and states that: An agent is a computer system that is situ-
ated in some environment, and that is capable of autonomous action in this
environment in order to meet its design objectives.
Focusing on the presented definition and considering other approaches
to agency, it has to be stated that autonomy is the main principle of an
agent. Another point of the presented definition indicates relation between
an agent and the environment in which this agent acts and percepts. An agent
K. Cetnarowicz et al.: An Immunological and an Ethically-Social Approach to Security Mech-
www.springerlink.com c© Springer-Verlag Berlin Heidelberg 2006anisms in a Multiagent System, Advances in Soft Computing 5, 11–19 (2006)
12 Krzysztof Cetnarowicz et al.
should be seen as a part of his environment especially from a point of view
of another agent functioning in this environment. “Design objectives” of an
agent which influence behavior of an agent are important here. Considering
security issues in an agent system it have to be stated that “design objectives”
of an autonomous agents might be hidden — not explicit shown to other
agents functioning in this system.
Several architectures for intelligent agents which facilitate creating of an
intelligent agent are proposed (e.g. logic-based architectures, reactive archi-
tectures, Belief-Desire-Intention architectures), but there is no explicit def-
inition how agents should be designed. An agent could be realized in the
form of a program, an object or just a function. Many ideas in the subject
of agency concentrate on the ability of an agent to reason on the base of his
knowledge or even learn from his experience. There is nothing stated about
the knowledge representation used to learning or just reasoning of an agent.
The environment of agents might be closed or open. Because of the in-
creasing interconnection and networking of computers, the most considered
environments are open in the sense of enable mobile agents to get in such an
environment and act inside it. An environment of agents should enable an
agent to interact with other agents existing in this environment. Agents of
different types which have different possibilities to percept, reason and act
might exist in an environment. Agents functioning in one environment might
have different objectives, goals which they want to achieve. Agents in such
a multiagent system should have communication and interaction possibilities
in order to achieve their goals. Considering mentioned definition of an agent
as a part of his environment, communication and interaction possibilities are
seen as possibilities to act and perceived.
Taking into consideration security issues in multiagent system it has to
be noticed that some agents functioning in an environment might have an
objective which is to make it impossible to act for other agents. This could be
achieved e.g. by exhaustion of some important resources in the environment,
flooding other agents by unnecessary messages etc. Another danger is the
possibility of creating an agent which might mislead other agents in order to
make actions which are not allowed to him.
2 Some Security Solutions in Multiagent Systems
Two solutions: the RETSINA infrastructure and the VIGIL system are pre-
sented briefly in order to show some security problems in multiagent environ-
ments. Wong and Sycara describe the design of a security infrastructure based
on RETSINA [12,13] — an open multiagent system that supports communi-
ties of heterogeneous agents. RETSINA contains two types of infrastructure
entities: ANSs (Agent Name Servers) and Matchmakers. An ANS maps an
agent ID to his address in the system. If an agent wants to contact with
another agent, whose address he does not know, he contacts with the near-
Security Mechanisms in a Multiagent System 13
est ANS. Matchmaker maps an agent ID to his capabilities. If an agent is
looking for another agent, which provides a particular service, he queries a
Matchmaker.
RETSINA guarantees agent authentication via a Certificate Authority.
Before starting up an agent, his deployer has to get the public key confirmed
by the Agent Certificate Authority. Only certificates of this type are accepted
in the system. The authors assumed that agents are able to decide whether
a certificate is valid. Agents may not accept a certificate as valid because
there may be multiple Certificate Authorities, not all of which are trusted
by all agents. For simplification they assumed a single Certificate Authority
in the system. Such solution forces the deployers of agents, which come from
outside the system, to perform the confirmation procedure again in a trusted
Certificate Authority otherwise these agents can not function in RETSINA
(they are not reachable by other agents, their services are not available and
their requests are rejected).
Another research on some approaches to the problem of security in multi-
agent system is presented in e.g. [7]. The authors present the system VIGIL
designed to provide security and access control in distributed systems. In this
system an agent is authorized to access a service if he possesses the required
credentials. The access decisions are also based on the roles that agent plays
as a part of an organization. The example presented by authors describe a
situation, in which a user in a meeting room is using a projector, he is prob-
ably the presenter and should be allowed to use the computer too. Therefore
rights can be assigned or revoked dynamically without changing an agent’s
role.
In VIGIL each new agent has to register in the Service Manager by sending
its digital certificate, a list of roles which he can access. The Service Manager
verifies the client’s certificate. This procedure ensures a trust relationship
between the Service Manager and the client. The agent’s certificate is also
sent to the Role Assignment Module, which decides about the roles the client
can have. If the set of roles is fixed, the client has all the rights associated with
them. If an agent wants to use some particular service he sends all required
credentials along with the request for service to the provider.
2.1 Conclusions of Presented Security Solutions
Presented solutions are closely related to the domain of usage. There could
be a multiagent system in which presented solutions are not proper. A reg-
istered agent could undertake actions that could be undesirable, for example
one registered agent could flood other agents with his messages. The pre-
sented procedures are in certain degree analogous to the mechanisms of user
authentication in a computer system. However, registered (certificated) user
could also undertake actions that are dangerous for the secured system.
Considering the example with a person using a projector and therefore he
should also be allowed to use a computer, some disadvantages of the VIGIL
14 Krzysztof Cetnarowicz et al.
system can be shown. A user (an agent) could use a computer in different
ways e.g. he could delete all applications and damage the operating system.
That is why presented security infrastructure does not prevent from abusing
resources that are in the environment of the secured system.
The authors of the RETSINA infrastructure present some security threats
in multiagent systems. To avoid the problems of corrupted naming and match-
making services they propose to use trusted ANSs and Matchmakers that
should service only valid requests. However, the authors do not present any
solution in case of ANSs or Matchmakers corruption, they just assume that
these elements are always trusted in the system. An ANS with corrupted
database or service could make some agents unreachable in the system or
flooded with requests for service that they do not provide. Analogous prob-
lems could occur when a Matchmaker becomes corrupted.
3 An Immunological Approach to Intrusion Detectionin a Multiagent System
The immune system defends the body against harmful foreign elements: cells
or molecules. In classical immunology this is the problem of distinguishing
molecules and cells of the body (called self ) from foreign ones (called nonself )
which should be eliminated. In [5,6,11,14] some ideas, questions and solutions
in the sphere of an immunological approach to detection of intruders in the
environment of computer system are presented. The principles of immune
system are described as attractive especially from the point of view of an
open system in which autonomous, distributed agents act.
In our discussion of an immunological approach to intrusion detection in
a multiagent system we would like to focus on the method based on the gen-
eration of T cells in the immune system. This method applied to a computer
system is described in e.g. [5,6,14]. Applying an immunological self–nonself
discrimination to the environment of an multiagent system involves two basic
problems: what should be processed by an immunological change detection
algorithm and what the collection of self strings S should contain.
3.1 Analyzed Strings in a Multiagent System
Strings that are analyzed by a change detection algorithm are fragments of
program code in the case of a computer system (as in [5]) or fragments of TCP
packets in the case of a network system (as in [6]). Analyzed strings describe
the activity of an element of computer system which is secured. Considering
a system with autonomous agents, it seems that fragments of agents’ code
are not suitable to recognize if activity of an agent is desirable or not in a
particular system. Reasons for this statement are:
• autonomy of an agent — possibility of changing activity,
• internal knowledge of an agent which could be modified.
Security Mechanisms in a Multiagent System 15
An autonomous agent might change his activity in order to some circum-
stances in the system which he is situated in. In some circumstances an agent
can present a desirable activity, which is the activity that will not cause the
damage of other agents or parts of the environment of the system which
is secured. However, the same agent can change his activity on the base of
his autonomously made decision. Reasons for changing this activity could
be various: from changing the circumstances in the environment to changing
activities of other agents.
Considering an agent which has an internal knowledge and which could
modify his knowledge, it seems that the code of an agent is related to the
knowledge that he possesses. The knowledge could be or could not be related
with activity of the agent possessing this knowledge. The knowledge of an
agent should not be evaluated in the meaning of distinguishing desirable or
undesirable activity, because the knowledge is not related to the objectives
or the goals of an agent.
Concluding two presented points in the discussion it has to be stated that
the only possible way to describe an activity of an agent is to present actions
that this agent undertakes. These actions form specific behavior of the agent.
3.2 Self/Nonself in a Multiagent System
The question how self strings should be chosen is a basic problem of ap-
plication of an immunological change detection algorithm. In the case of a
computer system there could be fragments of programs that are inside the
secured system in the time of detector generation. In the case of a multiagent
system this problem is not so obvious because of the dynamic character of
agents which can move from inside or outside of the system. The environment
of the multiagent system can include different agents or agents of different
type in a different time period. There is a possible scenario which can be
presented as:
1. there are 20 agents of type A and 10 agents of type B in the system,
2. the sets of detectors are generated,
3. all agents of type A and B move outside a system,
4. 10 new agents of type C appear in the system.
Now we have to consider which agents should be distinguished as self or
nonself. Assuming each type of agent has another behavior, nonself agents
should be an agent of type C. But it is possible that an agent of type B also
should be considered as nonself, because the goal of that agent is to destroy
the system.
The problem presented in mentioned simple example seems complicated,
because the openness of a multiagent system prevents from decision if all
agents inside a protected system could be considered as self and could be used
16 Krzysztof Cetnarowicz et al.
by the negative selection in the algorithm of creation of detectors. A creator
of a multiagent system doesn’t have any knowledge if there are or there aren’t
any agents inside a system which activity are dangerous. This deadlock in
our discussion can be broken by the use of ethically–social approach that is
presented in the next chapter.
3.3 The Danger Theory and an Immunological Approach
The theory of functioning of human immune system presented above is con-
tradictory to some experiments or observations, e.g. there is no immune reac-
tion to foreign bacteria in the gut, however this bacteria should be recognized
as a nonself element which is synonymous (in classical immunology) to the
intruder that should be killed. Because of many faults and inconsistency in
traditional self /nonself theory, a new Danger Theory was presented. This
new theory presented among others in [8,9] is still improved and still not
completed, but it enable to present the mechanisms which are not limited to
very restricted, and in many cases unreal, self /nonself discrimination. The
main idea in the Danger Theory is that the immune system does not respond
to nonself but to danger. The danger is measured by the damage done to cells
indicated by distress signals that are sent out when cells die an unnatural
death. Because distress signals are just results of harmful intruders behavior,
the most important issue in the Danger Theory is using the results of an
entity behavior in the process of an intruder detection.
The Danger Theory correspond with our idea of ethically-social approach
to security in multiagent systems, which one of two main assumptions is
the evaluation of behavior (or more precise evaluation of visible results of
behavior) as it is presented in Sect. 4. In current state of art of Artificial
Immune Systems many ideas about the applications of mechanisms occured
in the Danger Theory are now discussed. Some ideas presented in [1] are also
related to using some information about results of behavior of an intruder —
e.g. using of the signal sent in a case of too low or too high memory usage as
a result of infection. Such data should be used in the early stage of attack in
order to limit and minimize the damage.
4 Ethically-social Approach to Security Problem inMultiagent Systems
Considering the dynamic of a multiagent system environment, the problem
is which agents should be regarded as self or nonself (which is equal to
an intruder in classical immunology). Rapidly developing agent technology
makes the full flow of resources among open computer systems possible. Au-
tonomous agents can migrate in the net without knowledge of the owner
or an administrator. Agents can also execute their tasks without anybody’s
to Intrusion Detection
Security Mechanisms in a Multiagent System 17
knowledge. These tasks can be useful for the owner as well as destructive for
the system. It could be shown in an example in which two agents fulfill two
extremely different functions:
• as an intruder migrating and searching for system, which it can attack,
• as an agent, sent by a friendly system to improve protection.
Immunological self /nonself mechanisms (applied to code of agents) do
not appear useful to investigate the introduced case of two migrating agents.
Both mentioned agents would be classified by immunological system as non-self which means undesirable. Treatment of a friendly but nonself agent as
undesirable is not proper, what is also consistent with the newest Danger
Theory. It would be proper to divide computer resources into the following
parts:
• good – desirable agents in an environment in which they act,
• bad — undesirable.
Distinguishing between good and bad can be accomplished only on the
basis of the observation of work or intentions of acting resource. Such ob-
servation takes place in a small society of cooperating people. Each person
observes the behavior (actions which are undertaken) of all other people in
the society. As a result behavior of one person is observed by all entities of
the society. The decision if person A should cooperate with person B is made
not only on the basis of the evaluation of person B’s behavior which is made
by person A, but also opinions of other members in the society are consid-
ered. The opinion of the whole society about one particular person consists
of many opinions of all persons in the society. The ethically-social approach
to security problem induces two main problems in a multiagent system:
• how to design behavior evaluation mechanisms in which every agent in a
system should be equipped,
• how to collect and process distributed evaluations which are made by all
agents in a secured multiagent system.
4.1 Behavior Evaluation Mechanisms
The behavior of an agent determine actions which an agent undertook. These
actions should be seen as objects which create a sequence. The sequence of
actions could be registered by an agent which observes evaluated agent. The
registered objects–actions could be processed in order to qualify whether it
is a good or a bad acting agent.
A change detection algorithm could be used to the evaluation of the se-
quence of actions. The immunological intruders detection in the computer
environment has to be done on the basis of certain characteristic structures.
These structures in the case of behavior observation are chains of actions per-
formed by an observed agent. These chains are of the settled length l, so one