mieczysław a. kłopotek, sławomir t. wierzchon, krzysztof

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]

[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

““ “

“ “


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


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


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


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


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

[email protected]

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


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.


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


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


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

mieczysław a. kłopotek, sławomir t. wierzchon, krzysztof

Documents