VOLUME 25 NUMBER 1 2001
 
Abstracts:

Neural network analysis of time series data 
Ryszard Tadeusiewicz Pawel Lula
University of Mining and Metalurgy, Mickiewicza Av. 30, 30-059 Cracov, Poland  rtad@biocyb.ia.agh.edu.pl, neural@uci.agh.edu.pl 

Pawel Lula

Department of Computer Science, University of Economics, av. Rakowiecka 27, 31-510 Poland eilula@cyf-kr-edu.pl

Neural networks have properties known to be effective in the modeling of economic phenomena. The process of constructing neural models that represent one-dimensional time series is reviewed and demonstrated. Explicit attention is paid to the evaluation of the models. All steps required by theory and practice are demonstrated through an example. (pp. 3-10)

Keywords: neural models, neural networks


Applying neural networks to practical problems - methodological considerations 
Marcin Paprzycki, Aaron Costeines, William Douglas, Paul Gatling, Rick Niess and Lenny Scardino
Department of Computer Science and Statistics, University of Southern Mississippi, Hattiesburg, MS 39406-5106, USA m.paprzycki@usm.edu 

Results of performance comparison of seven neural network architectures applied to a simplified multifont recognition problem are used to illustrate the need for more comparative performance studies. This need is a result, among others, of increasing computational power of modern PC's and development of commercial neural network software packages that combine multiple NN architectures in one easy-to-use environment. Experimental data supporting our conjecture is presented and discussed. (pp. 11-17)

Keywords: neural networks, performance comparison, simplified multifont recognition problem


Artificial neural network approach to data analysis and parameter estimation in experimental spectroscopy 
Mehmed M. Kantardzic, Anna Goldenberg and Troy E. Howe
CECS Department, Speed Scientific School, University of Louisville, Louisville, KY 40292

Peter Faguy

Department of Chemistry, University of Louisville, Louisville, KY 40292

The P- and S-polarized Infrared Reflection Absorption Spectra (IRRAS) can be experimentally measured and, if the optical constants for the three phases are known, the film thickness can be determined. Film thickness measurements are crucial for many surface applications. We propose the methodology of thickness measurement based on artificial neural network (ANN) technology. Using retro modeling we reduce the dimensionality of the problem, and prepare training data set for an ANN. Simulation and 3D visualization based on proposed model help us to validate the robustness of the trained ANN, and to estimate unknown parameters of instrumentation in experimental spectroscopy. Experimental IRRAS data were the inputs for an ANN testing phase, where the output is the estimated value of the film thickness.

Keywords: neural networks, spectroscopy
 


Control of a one-legged three-dimensional hopping movement system with multi-layer-perceptron neural networks
Klaus D. Maier

Infineon Technologies AG, CMD CE, PO Box 800949, D-81609, Munich, Germany, k.maier@ieee.org

Volkmar Glauche, Clemens Beckstein

Friedrich-Schiller-Universitiy Jena, Institut of Computer Science, Artificial Intelligence Group, D-07740 Jena, Germany

Reinhard Blickhan

Friedrich-Schiller-Universitiy Jena, Institut of Sports Science, Biomechanics Group, D-07740 Jena, Germany

Controlling the model of a movement system based on the dynamics of biological hopping and running is investigated. This movement system consists merely of a massless spring attached to a point mass. It is describing fast three-dimensional legged locomotion on even grounds. Rapidly moving legged autonomous systems require different hardware layouts and control approaches in contrast to slow moving ones. The spring mass system is a model that describes this principle movement as well as the principle control task. Multi-layer-perceptrons (MLPs) were used to implement neurocontrollers suitable for such a movement system. They prove to be suitable for exact control of the movement with a relatively small number of neurons. This is also shown by an experiment where the environment of the spring-mass system has been changed from even to uneven ground. The neurocontroller is performing well with this additional complexity without being trained for it. (pp. 27-38)

Keywords: neural networks, performance comparison, simplified multifont recognition problem


Neuro-fuzzy synergism for planning the content in a web-based course
G.D. Magoulas
Department of Information Systems and Computing, Brunel University, UB8 3PH, United Kingdom George.Magoulas@brunel.ac.uk

K.A. Papanikolaou and M. Grigoriadou
Department of Informatics, University of Athens, T.Y.P.A. Buildings, GR-157 84 Athens, Greece, {spsp, gregor}@di.uoa.gr

In this paper, neuro--fuzzy synergism is suggested as a means to implement intelligent decision making for planning the content in a web--based course. In this context, the content of the lesson is dynamically adapted to the learner's knowledge goals and level of expertise on the domain concepts s/he has already studied. Several issues that affect the effectiveness of the lesson adaptation scheme are investigated: the development of the educational material, the structure of the domain knowledge and the assessment of the learner under uncertainty. A connectionist--based structure is proposed for representing the domain knowledge and inferring the planning strategy for generating the lesson presentation from pieces of educational material. The learner's assessment is based on relating learner's behavior to appropriate knowledge and cognitive characterisations and on embedding the knowledge of the tutors on the learning and assessment processes into the system by defining appropriate fuzzy sets. The proposed neuro--fuzzy adaptation scheme is applied to a web--based learning environment to evaluate its behavior and performance. (pp. 39-48)
 

Keywords: adaptive learning environments, web-based course, adaptive lesson presentation, neural networks, fuzzy logic, leaner evaluation, reasoning under uncernatinty


Risk logical and probabilistic models in business and identification of risk models 
E.D. Solojentsev, V.V. Karasev
Institute of the Problems of Mechanical Engineering, Russian Academy of Sciences, Bolshoy pr. 61, St. Petersburg, 199178 Russia, sol@sapr.ipme.ru

The apparatus of logical and probabilistic (LP) simulation, not very popular among mathematicians and economists, is developed and used to study risk in business. The logical operations ( AND, OR, NOT ) are applied to the initiating events (instead of the traditional arithmetical addition of values). We present statistical data about risk objects as a table of "object-signs". A risk object is described by a large number of signs, and every sign has up to 10 gradations. Events-signs are connected logically and event-gradations are treated as groups of incompatible events (GIE). The events are given clear probabilistic sense. Risk LP-models for banks, business, insurance and quality are built as associative models on the basis of common sense and are considered as the hypotheses. (pp. 49-55)

Keywords: logic, probability, risk, model, identification, business


Performance analysis of a parallel neural network training code for control of dynamical systems
Javier E. Vitela, José L. Gordillo
Instituto de Ciencias Nucleares, Universidad Nacional Autónoma de México, 04510 México D.F., México

Ulf. R. Hanebutte
Center for Appiied Scientific Computing, Lawrence Livermore National Laboratory, Livermore CA 94551, USA

In this work we present the performance analysis of a parallel neural network controller training code that uses MPI, a standard portable message passing environment. The physical system used in this study is a nonlinear model of a tokamak thermonuclear fusion reactor in which the parameters were taken from the CDA-ITER (Conceptual Design Activity) group. A SGI/Cray Origin 2000 multiprocessor platform was used in the comprehensive performance analysis of the parallel code reported here, which includes a comparison of actual measurements with the results of the theoretical performance of ideal models. Three different assigments schemes for load balance were used in this study. (pp. 57-67)

Keywords: Parallel computing, artificial neural networks, nonlinear control, fusion reactor, message passing, MPI, SGI/Cray Origin 2000


Fault-tolerant ATM switching architectures based on MINs: A survey
Mohsen Guizani
Computer Science Department, University of West Florida, mguizani@cs.uwf

Muhammad Anan
Broadband Network Management Development, Sprint Telecommunications Company, muhammad.anan@mail.sprint.com

Multistage Interconnection Networks (MINs) offer an attractive way of implementing fast packet switching communication networks. Due to providing a very large capacity, MINs play an important role in building ATM switches. With the throughput requirements of the packet switches exceeding several Gigabits/sec, it is important to improve the fault-tolerance of these switches. The specific fault-tolerance methods used in the system plays an important role in the availability, reliability, and performance of the overall system. In this paper, a survey of fault-tolerant ATM switch fabric architectures is presented. The intention is to give a descriptive overview of the major techniques used to achieve fault-tolerant ATM architecture. The survey covers many important techniques that have been proposed for fault tolerant MINs. It discusses the fault-tolerace approach for computing systems reliability. In addition, it goes through some quantitative techniques to evaluate fault-tolerant systems. (pp. 69-81)

Keywords: MIN, fault-tolerance, ATM switch, banyan network, SEROS, self-routing, reliability, availability


Soft computing on small data sets
Bojan Novak
University of Maribor, Faculty of Electrical Engineering, Computer Science and Informatics, Smetanova 17, 2000 Maribor, novakb@uni-mb.si

The fusion of artificial neural networks (ANN) with soft computing enables to construct learning machines that are superior compared to classical ANN because knowledge can be extracted and explained in the form of simple rules. If the data sets are small it is hard to find the optimal structure of ANN because classical statistical laws do not apply. One possible remedy is the structural risk minimization method applied together with a VC dimension estimation technique. The construction of the optimal ANN structure is done in the higher dimensional space. The distortion of an image in this transformation can happen and the widely used expression for VC estimations based on minimal input data enclosing hypersphere and margin is not precise. An improvement of VC dimension estimation is presented. It enables better actual error estimation and is particularly suitable for the small data sets. Tests on some real life data sets have confirmed the theoretical expectations. (pp. 83-88)

Keywords: soft computing, machine learning, neural networks
 


Computerized logistics information systems - a key to competitiveness

Anton Cizman

University of Maribor, Faculty of Organizational Sciences, Kodriceva 55/a, 4000 Kranj, Slovenia, anton.cizman@fov.uni-mb.si

Part of an organization's ability to use logistics as a competitive weapon is based on its ability to assess and adjust actual logistics performance real time. This means the ability to monitor customer demands and inventory levels as they occur, to act in timely manner to prevent stockouts, and communicate potential problems to customers. This requires excellent, integrated logistics systems which impact all of the logistics activities. In this paper we examined how computer and information technology can be used to support logistics management. Customer order cycle and order processing systems are pointed out first. Then advanced information technologies such as decision support systems, artificial intelligence, and expert systems which are being used directly to support decision making in logistics, are examined. (pp. 89-98)

Keywords: information systems, logistics management, oreder processing, decision support systems, information tecnology, manufactoring, services, order processing


An agent that understands job descriptions

Andraz Bezek and Matjaz Gams
Jozef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia, andraz.bezek@ijs.si , matjaz.gams@ijs.si 

An important property of intelligent agents is semi-understanding of desired tasks. We have designed an agent module referred to as Professional Description Assistant that is able to identify the corresponding job definition from a job description in textual form. Thus it plays a significant role in the advanced services of our EMA employment agent. By means of some standard and modified learning methods the findings proved to be quite encouraging. The new function aids the user performing some standard tasks, which normally demand human decision and experience, to a great extent. (pp. 99-105)

Keywords: intelligent agents, understanding, intelligent services


A nested combinatorial-ctates model of sentence processing

Arthur C. Brett, Tadao Miyamoto and Joseph F. Kess                                                                                                                                 Department of Linguistics, University of Victoria, Victoria, British Columbia, Canada V9W 3P4

Natural languages differ in the directionality of their syntactic build-up, developing from the phrasal head or matrix clause in either a left-branching direction or a right-branching direction. Japanese, for example, is quite consistently left-branching, while English is generally right-branching. We argue that the purported greater complexity of left-branching sentences arises from a confusion of analytical syntactic structure with sentence processing mechanisms, and we propose a model based upon the Chalmers' Combinatorial-State Automaton (CSA) which we suggest more nearly approximates actual neural processes. Our model, a Nested Combinatorial-States Automaton (NCA), can be constrained to require only finite memory resources, and assumes that the greater part of sentence processing occurs at the lexical level. (pp. 107-122)

Keywords: natural language processing, psycholinguistics, finite automata


Petri nets and IDEF diagrams: applicability and efficacy for business process modelling

Vesna Bosilj - Vuksic                                                                                                                                                                                  University of Zagreb, Faculty of Economics, Department of Business Computing, Trg J.F. Kennedya, 10000 Zagreb, Croatia, vbosilj@efzg.hr

Vlatka Hlupic                                                                                                                                                                                                  Brunel University, Department of Information Systems and Computing, Uxbridge, Middlesex UB8 3PH, United Kingdom, Vlatka.Hlupic@brunel.ac.uk 

It is apparent that developing dynamic models of business processes prior to their radical change could increase the success of BPR projects. This paper investigates a suitability of IDEF diagrams and Petri Nets for modelling business processes. Information modelling and simulation modelling are discussed from the business process re-engineering perspective. Examples of business process modelling using IDEF diagrams and Petri nets are presented. The suitability of these two graphical methods for business process modelling is discussed, and a comparison of usage of these two methods for BPR is provided. (pp. 123-133)

Keywords: business process reengineering (BPR), business process modelling, simulation modelling, IDEF0 diagrams, IDEF3 diagrams, Petri nets, DES-nets


The object model of splines
Mojca Indihar Stemberger and Janez Grad                                                                                                                                                         University of Ljubljana, Faculty of Economics, Kardeljeva ploscad 17, SI-1000 Ljubljana, Slovenia, mojca.stemberger@uni-lj.si, janez.grad@uni-lj.si 

Data analysis like Data Mining is getting more and more important every day because the organizations want to get their competitive advantage. Splines are piecewise polynomial functions that are used for the analysis by some Data Mining tools. Except that, they are widely used in CAGD applications for designing. The article presents the object model of splines that is independent of the concrete application but can serve as the basis for any application which uses splines and reduces the gap among such applications and their incompatibility. It was modeled by object-oriented modeling language UML. We also propose how to save splines in an ODMG compliant database. The prototype of a system was developed in Java and is presented in the article as well. (pp. 135-142)

Keywords: splines, object-oriented modelling, object-oriented database, UML, ODMG, Java