Abstracts:
A Safe and Efficient Agent Architecture
Soe-Tsyr Yuan
Information Management Department, Fu-Jen University,
No. 510 Rd. Chung-Chen, Sing-Zoung, Taipei, Taiwan, R.O.C;
yuans@tpts1.seed.net.tw
When an agent pursues human's goals, the agent should generate an acceptable solution (accordingly protecting humans from harm). We call this necessity safety enforcement. Furthermore, the agent should exercise safety enforcement in a computation tractable manner. In a complex world, the agent often does not have omplete information. How can an agent maintain its safety enforcement with incomplete information in a computation tractable manner? This paper presents an agent architecture which can be used to solve the above problem. This framework combines the culture support in the domain ontology and the decomposition and causality of partial-order planning. (pp. 161 - 171)
Keywords: software agent, safety enforcement, decomposition and causality of partial-order planning, ontology
Multi-Agent Systems as a Paradigm for Intelligent System Design
Amal El Fallah Seghrouchni
LIPN - CNRS / URA- 1507,
Institut Galilee University Paris13- FRANCE;
elfallah@lipn.univ-paris13.fr
The design of complex and intelligent systems requires an adequate framework that should combine the rigor of formal models, the practicality of existing development methods and the performance analysis of modeling tools. Based on the Multi-Agent System paradigm, this paper proposes an efficient approach to build intelligent systems, the active components of which are cognitive agents (i.e. an agent is endowed with various abilities to solve problems, interact with other agents). In our approach, agent behavior is viewed as an organized collection of abstract actions considered as an abstract control pattern. As agents are willing to cooperate, they exchange their respective plans and coordinate them to cancel negative interactions and to take advantage of helpful ones. The coordination mechanism relies on distributed planning. Concurrent plans are specified through the Recursive Petri Net formalism. A transformation method refines an agent's plan (i.e. control pattern) dynamically thus enabling agents' roles to be predicted and validated formally. The coordination model combines two paradigms: remote and situated coordination that satisfie the following requirements: domain independence, broad coverage of interacting situations, operational coordination semantics and natural expression for the designer.(pp. 173-184)
Keywords: collaborative problem solving, coordination, distributed planning, interaction, concurrency, multi-agent system, recursive petri nets
Internet Information Brokering: A Re-Configurable Database Navigation, Data Filter and Export System
Syed Sibte Raza Abidi
School of Computer Science, Universiti Sains
Malaysia,
Penang, Malaysia;
sraza@cs.usm.my
The abundant use of the Internet has promoted it as an effective medium for global information sharing, for instance accessing information stored in remote databases that are inter-linked by the Internet. We present an Internet-based database application, a 'Re-configurable Internet Information Broker' (RIIB) that provides (a) the functionality to dynamically connect and interact with remote databases via the Internet; and (b) an 'intelligent' database navigation engine which is based on the notion of database virtual hierarchies - a database navigation mechanism that allows users to navigate the database by envisaging it as a user-defined hierarchical structure.(pp. 185-191)
Keywords: information broker, database navigation, database virtual hierarchy
MANIS: A Multi-Agent System for Network Information Services
Xiaoqiang Kang
Department of Computer Science and Technology
Tsinghua University, Beijing, 100084, China;
kangxq@public.bta.net.cn
Chunyi Shi
Department of Computer Science and Technology
Tsinghua University, Beijing, 100084, China
Network information service has been an important ground for distributed AI research. In this paper, we present a multi-agent system, called MANIS, which integrates several methods for text understanding and offers the text understanding services over network. MANIS is organized as a market-like system. Broker agents can choose among a variety of services offered by server agents based on the sharing information in the service market. On the other hand, server agents can reorganize themselves via decomposition and composition in order to increase the utility of their limited computational resources. MANIS provides an open framework for building multi-agent systems for network information services and will serve as a testbed for our future researches.(pp. 193-200)
Keywords: distributed AI, multi-agent systems, network information service
Cognitive Simulation of Operator's Diagnostic Strategies in Nuclear Power Plants
Se Woo Cheon
Korea Atomic Energy Research Institute Yu Song P.O. Box 105, Taejon 305-600,
Korea;
swcheon@nanum.kaeri.re.kr
Jeong Woon Lee, and Bong Sik Sim
Korea Atomic Energy Research Institute Yu Song P.O. Box 105, Taejon 305-600,
Korea
Jin Kyun Park and Soon Heung Chang
Korea Advanced Institute of Science and Technology,
Kusong Dong 373-1, Yu Song Gu, Taejon 305-701, Korea
This paper describes an approach to simulate operator's diagnostic strategies under emergency situation of nuclear power plants. To assess operator's cognitive workload during the diagnosis, we have developed a diagnostic model, which is an embedded module of a task simulation analyzer, SACOM. This model is based on a blackboard architecture and uses four types of diagnostic agents. The diagnostic agents can simulate typical diagnostic strategies used by human operators: i.e., data-driven search, symptomatic search, hypothesis-driven search, and topographic search. The model is developed using a G2 expert system tool.(pp. 201-207)
Keywords: cognitive simulation, diagnostic strategies, blackboard architecture, nuclear power plants
Modelling Human Cognition in Problem Diagnosis: A Hybrid Case-Based Reasoning Methodology
Yuh Foong David Law
National University of Singapore, Singapore 119260;
mcblawyf@leonis.nus.sg
As modern business functions become more knowledge-intensive, the demand for intelligent and knowledge-based systems has increased. Part of the challenge of developing quality expert systems lies in the effective modelling of cognitive processes of human experts and representation of various forms of related knowledge in a domain. This paper describes a cognitive-driven methodology for the development of expert systems based on a hybrid reasoning approach. The methodology incorporates a hybrid case-based reasoning (CBR) framework of techniques which include case memory organisation networks, case indexing and retrieval schemes; and an interactive and incremental style of reasoning. The paper will discuss the design and implementation of an expert system in a problem response help desk environment of a local bank, as an example of a successful application of this methodology. (pp. 209-217)
Keywords: hybrid case-based reasoning, cognitive processes, fault diagnosis, help desk
Medical Decision Support System for the Management of Hypertension
Young Moon Chae
Department of Health Management Information System, Yonsei University,
C.P.O.Box 8044, Seoul Korea;
ymchae@hitel.kol.co.kr
Seung Hee Ho, and Mi Young Bae
Department of Health Management Information System, Yonsei University,
C.P.O.Box 8044, Seoul Korea
Sang Hyun Lee
Department of Family Medicine, Inha University
Hee Choul Ohrr
Department of Preventive Medicine, Yonsei University
This paper describes the development of a medical decision support system (MDSS) for the management of essential hypertension. While most MDSS's have been developed for doctors in university hospitals, this system is developed for primary health care doctors and health workers in community health centers. The MDSS consists of three modules: health management information system (HMIS) for health center, logistic regression model, and an expert system. HMIS maintains hypertension database for two modules. Logistic regression was used to identify risk factors for essential hypertension and three knowledge models were compared to determine the best method for predicting the severity of hypertension and the prognosis of treatment. Out of all the cases, discriminant analysis showed the highest predicitive power. To improve user interface, computer assisted instruction (CAI) capability also incorporated in the system.(pp. 219-225)
Keywords: hypertension, decision support system, expert system, CAI (computer assisted instruction), neural network, case-based reasoning, discriminant analysis
Lattice-based Knowledge Discovery in Network Management Data
F.J. Venter
Department of Computer Science,
University of Pretoria, 0002, South Africa;
fventer@cs.up.ac.za
G.D. Oosthuizen
Department of Computer Science,
University of Pretoria, 0002, South Africa;
goosthui@cs.up.ac.za
J.D. Roos
Department of Computer Science,
University of Pretoria, 0002, South Africa;
jroos@cs.up.ac.za
Complex and large systems can be hard to manage if the elements that they comprise have dynamic causal relationships. Management of such a system usually entails monitoring and interpreting data generated by various elements of the systems. This data needs to be processed into meaningful information, so that a manager can use the information to get enough insight into the dynamic state of the system to be able to proactively influence the run-time behavior of the system. However, if the amount of data generated for a system is large enough and the complexity of the relationships between elements of the system is high enough, then traditional information analysis and presentation methods do not suffice to give the manager of the system a comprehensive understanding of the state of the system. What is needed is more intelligent analysis of the possibly enormous corpuses of data streaming to the management system in real time. The field that is well positioned to give this sort of analysis is Knowledge Discovery in Databases and Data Mining (KDD) - a process that has the goal to extract knowledge from data. A range of techniques, including neural networks, rule-based systems, case-based reasoning, machine learning, statistics, etc. can be applied to the problem. We will discuss a KDD approach that employs a concept lattice to determine dependencies between elements of a computer network. We will firstly discuss the problem form a network management (NM) point of view and then show how KDD and specifically lattice-based KDD will address the problem(pp. 227-238)
Keywords: knowledge discovery, data mining, network management, concept lattice
Hiroshi Hatakama
Fujitsu Limited, 2-15-16 Shin-yokahama, Kouhoku-ku,
Yokohama 222, Japan;
hatakama@sysrap.cs.fujitsu.co.jp
Takao Terano
Graduate School of Systems Management, University
of Tsukuba 3-29-1 Otsuka, Bunkyo-ku, Tokyo 112, Japan;
terano@gssm.otsuka.tsukuba.ac.jp
This paper proposes a knowledge intensive learning model in a multiagent system, based on the concepts described in organizational learning of management science literature. We examine the validity and feasibility of the model in order to apply it to distributed heterogeneous knowledge systems. We have implemented two typical variations of the model: the Specialists-Model and the Generalists- Model. Using the two variations, we carry out simulation studies on the dynamic behaviors of decision making and learning in the organization of agents. The experiments showed that the model had unique adaptability to the change of environments, by enabling agents to efficiently supplement knowledge each other.(pp. 239-247)
Keywords: multiagent model, organizational sciences, explanation-based learning (EBL)
A Case Study in Non-monotonic Reasoning: an Alternative Analysis of the Yale Shooting Problem
Yong Sun
Department of Computer Science,
The Queen's University of Belfast, Belfast BT7 1NN, Northern Ireland;
Y.Sun@qub.ac.uk
Non-monotonic reasoning is often used in daily life, e.g. lacking evidence to the contrary of beta we infer beta. This kind of logic has its application in queries to knowledge bases, very large databases and deductive bases. This case study uses the "Yale Shooting" problem to present non-monotonic reasoning from a proof-theoretic standpoint. Basic knowledge of first order logic and many-sorted algebras is assumed.(pp. 249-251)
Keywords: non-monotonic reasoning, Yale shooting problem
The Weighting Issue in Fuzzy Logic
Xudong Luo and Chengqi Zhang
Department of Mathematics,
Statistics and Computing Science, The University of New England,
Armidale, NSW 2351, Australia
Jingqiu Cai
Department of Computer Science,
Xiamen University, Fujian 361005, China
The objective of this paper is to establish a class of models for solving the weight problem in fuzzy logic. First, some constraints that weighted fuzzy logic should satisfy are given based on Triangular norms and conorms. Then, based on these constraints, a class of models (referred to as relative weighted models) are established for handling weights in fuzzy logic. These models are novel in three aspects: (1) they include non-weighted models as their special cases, (2) the weighted conjunction and weighted disjunction can be distinguished from each other, and (3) the information from all sub-propositions can be sufficiently considered. In addition, this paper proposes principles for selecting proper weighted models.(pp. 255-2622)
Keywords: fuzzy logic, relative weighted models, uncertainty, weight, triangular norms, triangular conorms
Principle-based Parsing and Logic Programming
Matthew W. Crocker
Centre for Cognitive Science, University of Edinburgh,
2 Buccleuch Place, Edinburgh, UK EH8 9LW
While deductive parsing techniques are well-understood for traditional rule-based and lexicalist grammars, they are rather more elusive for current principle-based grammars. In this paper, we argue that a major source of difficulty arises from a fundamental difference in the way such grammars should be axiomatised. While rule-based grammars typically consist of a set of sufficient 'structure-generating' axioms, principle-based grammars are more naturally expressed as a set of necessary 'structure-licensing' conditions. On this basis we propose a methodology for implementing deductive parsers which is more suitable for this class of 'licensing grammars'. We then argue that current principle-based grammatical theories can be most naturally implemented by decomposing them into representationally homogeneous subsystems, which are axiomatised as licensing 'sub-grammars' - each contributing its own aspect of a global syntactic deduction system. Finally we consider the new range of options this approach offers for developing flexible and possibly distributed control regimes.(pp. 263-277)
Keywords: natural language processing, computational linguistics, deductive parsing
Keqin Li
Department of Mathematics and Computer Science,
State University of New York, New Paltz, New York 12561-2499, USA;
li@mcs.newpaltz.edu
Yi Pan
Department of Computer Science, University of Dayton,
Dayton, Ohio 45469-2160, USA
A parallel system is a parallel algorithm together with a parallel machine which supports its implementation. Scalability of parallel computing systems describes a property which exhibits performance linearly proportional to the number of processors employed. This paper is devoted to the investigation of average-case scalability of random parallel algorithms executing on multicomputer systems.Such probabilistic analysis is important in revealing the average-case performance of parallel processing, and performance of randomized parallel computations. We study the effect of communication overhead on speedup, efficiency, and scalability of such parallel systems. In particular, we characterize the communication overhead such that the expected efficiency can be kept at certain constant level, and that the number of tasks grows at the rate Theta(Plog P), where P is the number of processors. We also show that due to load imbalance, Theta (Plog P) scalability of the number of tasks is the best possible in our probabilistic model of parallel computations. Task granularity and isoefficiency functions are also discussed. (pp. 279-291)
Keywords: bus connected system, communication cost, completely connected system, efficiency, execution time, performance analysis, random parallel program, scalability, speedup
Algorithms in the Method of Paired Comparisons
George J. Miel and P. Diane Turnbough
University of Nevada, Las Vegas,
NV 89154, U.S.A;
miel@cs.unlv.edu
A description is given of a computational model for ranking elements of a given set based on pairwise comparisons of the elements when the set is large and the comparisons are unstructured. The model involves a large sparse overdetermined linear system Cx=d, where C is the mxn incidence matrix of a graph with n nodes (elements of the set) and m arcs (paired comparisons) and where d is the vector of observed differences in worth. Under the assumption that the graph has q connected components, simple algorithms are given for computing efficiently the corresponding least squares estimation in terms of a maximum of q nonsingular dense systems the sum of whose dimensions is bounded by n.(pp. 293-297)
Keywords: rank deficient least squares, graphs, pairwise comparisons, psychometrics
Development of Human Self in Novel Environment
A. Dreimanis
Environmental State Inspectorate Riga, Latvia
A synergetic analysis of the human Self - environment interaction is proposed from the viewpoint of the necessary development of human capacity of his harmonious involvement and viability in novel environment. The concept "environment", being treated in line of the Popper's and Eccle's thesis of human's three worlds would include a multitude of physical, ecological, mental, socio-cultural, economical, technological and other factors. According to the Ashby principle of requisite variety, the necessary condition of human's successful survival and viability in the more and more complex, novel and varying environment will be predominance of human's inherent variety over environmental variety. The primary source of growth of human's inherent variety - information and its organized form - knowledge. Mutual interrelations between information, adaptation and self- organization are specified. The importance of creativity as information generation source in evolving systems is emphasized. The roles of self-creativity and flexibility are considered in the context of interrelations of active and passive adaptation. A possible synergetical mechanism of creativity is deduced, and the probable routes of development of human's creative potential and creative approach to the problem solving are pointed out. The beneficial effect of moderate chaos in adaptation and developmental routes is emphasized.(pp. 299-303)
Keywords: environment, adaptation, self-organization, variety, creativity, chaos
Information Conundrum: Semantics... with a Payoff!
Jacek Kryt
Retired from Ryerson Polytechnic University. Toronto, Canada.
res.: address: 21 Heathcote Avenue, Willowdale ON, M2L 1Y6, Canada;
jkryt@acs.ryerson.ca
The sorry state of business information systems is no longer a secret. Since 1968 we have had a computer science curriculum which is supposed to provide us with properly trained graduates to handle the problems related to this new technology. But there is criticism of those graduates voiced from various quarters. They have been very helpful developing this new technology, but they have failed to satisfy the needs of those who in search for information expected to benefit from the flow of data facilitated by computers. This prompts the question: what is this information anyway? Computer scientists assert that information is just data processed by computers. I will show that this view is unacceptable as it causes considerable harm to information systems now, and is likely to have a negative impact on the information society. We need a clear understanding of the term "information" the most fundamental term for the Information Age.(pp. 305-319)
Keywords: computer science, curriculum, data, end user, information, information highway, information systems, requirements, systems failures