WIS'95 - Abstracts

These are people who delivered their papers already:
Jan T. Bialasiewicz
Janusz Blachowicz
Urszula Boryczka
Krzysztof Cetnarowicz
Pawel Cichosz
Krzysztof Ciupke
Grzegorz Dobrowolski
Jerzy W. Grzymala-Busse
Joel D. Gunn
Zbigniew Kowalkowski
Marek Kretowski
Jacek Maitan
Jacek Malko
Zbigniew Michalewicz
Ryszard Michalski
Henryk Mikol^ajczak
Wojciech Moczulski
Agnieszka Mykowiecka
Adam Mrozek
Jan Mulawka
Edward Nawarecki
Daryl O'Rourke
Leszek Plonka
Z. Ras
Franciszek Seredynski
Wl^odzimierz Skorupski
Andrzej Skowron
Roman Slowinski
Jerzy Stefanowski
Jaroslaw Stepaniuk
Zbigniew Swiatnicki
Wieslaw Szczesny
Wieslaw Traczyk
Roman Wantoch-Rekowski
Ryszard Winiarczyk
Jakub Wroblewski
Wojciech Ziarko
Jan M. Zytkow
Robert Zembowicz

Jan T. Bialasiewicz : (see LA)
On pattern clustering in probabilistic neural networks
Abstract: In this paper, the need for clustering training set patterns in probabilistic neural networks has been analyzed. It has been proposed to solve this clustering problem using -sufficient data reduction algorithms.
Keywords:
Urszula Boryczka : (see LA)
The Ant System
Abstract: We show how the Travelling Salesman Problem ( TSP ), as the one of problems from the NPC class, may be solved using the Ant--system, the new heuristic algorithm, in which many simple, loosely interacting agents collaborate in finding a good solution. We present the formal description of the Ant--system for the TSP. We also report the best set of parameters' values obtained in our experiments.
Keywords: ant system, travelling salesman problem, multiagent system, parallel algorithm
Pawel Cichosz and Jan J. Mulawka : (see LA)
Towards More Efficient Intelligent Agents Learning from Delayed Rewards
Abstract: Reinforcement learning provides an attractive framework for developing intelligent adaptive agents. The current research in this field concentrates on scaling reinforcement learning algorithms so as to make them applicable to complex real-world domains. In this paper we discuss two possible approaches to this problem: integrating reinforcement learning with planning using environment models and hierarchical reinforcement learning architectures. While these two ideas have been already investigated by several authors, we suggest here their novel and potentially more successful instantiations. In particular, we show how integrated learning and planning systems can be built using the TTD procedure, a recently proposed speed-up technique for reinforcement learning algorithms based on the methods of temporal differences, and we propose a sketch of a hierarchical conceptual reinforcement learning architecture, in which individual reinforcement learners learn decision policies for some simple behaviors, whose applicability conditions are represented and learned using symbolic concept learning methods.
Keywords: intelligent agents, reinforcement learning, temporal differences, planning, hierarchical learning
Krzysztof Ciupke : (see LA)
METHOD OF ESTIMATING THE QUALITY OF COMPOUND PRODUCT BASED ON ROUGH MEREOLOGY
Abstract: Product quality, in particular the problem of quality control and method of estimating the quality is one of the problems that every manufacturer has to solve. To manage this problem some tools are needed. The author suggested, that rough mereology could be applied for this. This paper presents an example of the method of estimating the quality of drop forging hammers which apply scheme of mereological connections.
Keywords:
Marek Druzdzel : (see LA)

Abstract:
Keywords:
Jerzy W. Grzymala-Busse and Joel D. Gunn : (see LA)
Global Temperature Analysis Based on the Rule Induction System LERS
Abstract: This paper describes the process of developing a rule-based model of the earth global temperature. Rules were induced by the system LERS (Learning from Examples based on Rough Sets). Emphasis is placed on two important techniques: discretization of numerical data and classification of unseen cases. Results of validation of different induction methods are also presented.
Keywords: Rule Induction, Discretization, Classification, Global Temperature, Leaving-One-Out.
Zbigniew Kowalkowski and Janusz Blachowicz : (see LA)
The Usage of Expert Systems For Diagnosing and Servicing Of Board Equipments
Abstract: PDC’s expert system shell, ESTA, is easy to use and a stand-alone environment for constructing advisory and decision support systems. ESTA runs under MS-WINDOWS 3.1 and it can be easily interfaced to 3rd- party applications like spreadsheets, databases, word processors, etc. via the built-in DDE-interface. ESTA includes an interface to PDC Prolog, which means that you can go beyond the built-in functionality - you can build your own extensions to ESTA or integrate ESTA into an existing PDC Prolog application. In addition to advisory systems, ESTA is a prototyping tool. A knowledge base in ESTA has a hierarchical structure from which a tree display is automatically drawn. You can edit or expand your knowledge base directly from the tree, giving an overview of your knowledge. The expert system presented in this paper assists the specialists in diagnosing problems about flaying control system SOS-3M.
Keywords:
Jacek Malko, Henryk Mikol^ajczak, Wl^odzimierz Skorupski : (see LA)
Advances in Artificial Neural Network Application to Electric Load Forecasting. Case Study: Power System in Poland
Abstract: Following previous publication of the authors \cite{cite:1} paper contains presentation of some advances in the field. Paper focuses on layered ANN as a predictor for short- and long-term electric load forecasting and on some nonstandard approaches to short-term prediction: Kohonen NN as an autoassociative memory and Hecht--Nielsen NN as a model--predictor. Examples of application such a models are presented on the basis of Polish power system data.
Keywords: power system, electric load, forecasting, artificial neural network, application.
Zbigniew Michalewicz : (see LA)
Do Not Kill Uneasible Individuals.
Abstract:
Keywords:
Ryszard Michalski : (see LA)
Inferential Theory of Learning as a Framework for Understanding the Nature of Learning Processes.
Abstract:
Keywords:
Wojciech Moczulski : (see LA)
Example Of Design Knowledge Acquisition Using Induction
Abstract: An example of application of inductive Machine Learning methods to identification of design knowledge on anti-friction bearing systems is presented. A problem domain is connected with design (thought of as selecting of design features) of assemblies and with completion of subassemblies with respect to a set of criteria and input data (brief foredesign). A method of representing design features of elements is presented. A way of collecting training and test events is described and exemplary data are shown. To identify rules the AQ-15 learning-by-induction program has been applied. The results obtained up-to-date and conclusions draught are briefly discussed. The rules identified are applicable in building knowledge bases of expert systems that may be used to automate or aid some stages of solving design problems of machinery.
Keywords: Machine Learning, Design Knowledge Acquisition, Induction, Rules identification, Training and Testing Events, Accuracy of Classification
Agnieszka Mykowiecka, .... : (see LA)

Abstract:
Keywords:
Edward Nawarecki, Krzysztof Cetnarowicz, Grzegorz Dobrowolski (see LA)
Informacyjne aspekty konstrukcji system\'ow agentowych
Abstract: W niniejszym artykule przedstawiono pewna^ pr\'obe^ formalnego opisu systemu agentowego, uwzgle^dniaja^cego wyste^puja^ce tam procesy informacyjne oparte na dw\'och grupach informacji. \par Pierwsza^ grupe^ stanowia^ informacje pozyskiwane przez agenta w spos\'ob "bierny", tj. w wyniku jego w\l{}asnych obserwacji \'srodowiska, druga grupa jest niezbe^dna, gdy wyste^puja^ sytuacje konfliktowe, wzgle^dnie istnieje potrzeba wsp\'o\l{}dzia\l{}ania pomie^dzy agentami. \par Prowadzone rozwa\.zania zilustrowane be^da^ przyk\l{}adami pilotowych rozwia^za\'n sys\-te\-m\'ow agentowych, realizowanych w Katedrze Informatyki AGH oraz wynikami uzyskanymi na drodze bada\'n symulacyjnych.
Keywords: inteligentne systemy zdecentralizowane, autonomiczny agent, wy\-miana informacji
Daryl O'Rourke : (see LA)
A Review of the State of Commercial AI Based Applications In The USA
Abstract: AI is not confined to the research and academic realm but is being used in numerous commercial, business, government and scientific applications. There is a substantial commitment to the growth and expansion of applied AI throughout all sectors of American technology companies. AI will soon expand into all levels of applications development and will integrate much of the conceptual research into even the most basic commercial applications.
Keywords:
Leszek Plonka, Adam Mrozek, Ryszard Winiarczyk and Jacek Maitan : (see LA)
Implementing Rule-Oriented Knowledge Bases on Smart Networks
Abstract: Knowledge-based systems utilize expert knowledge to solve problems normally requiring human intelligence. They are often applied in process control when it is difficult or impossible to use classical model-based approaches. Networks of smart sensors and actuators are the most recent trend in automatic control. In this approach, low cost embedded controllers are placed in the same package with sensors and actuators and connected by a network, so a central computer is not needed to execute the control algorithm. Decomposition of the control algorithm into distributed smart devices is one of the key problems. The paper demonstrates how the control algorithm can be represented by a rule-oriented knowledge base and how the knowledge base can be decomposed and implemented on a network of smart sensors.
Keywords: knowledge-based systems, process control, rule-based systems, smart networks, state transition table
Zbigniew W. Ras (see LA)
Cooperative Query Answering
Abstract: Traditional query processing provides exact answers to queries. It usually requires that users fully understand the database structure and content to issue a query. Due to the complexity of the database applications, incorrect queries are frequently posed and the users often receive no answers or they might need more information than they have received. Cooperative query answering analyzes the intent of the query and provides generalized, approximate, or associated information that is related to the query.
Keywords:
Franciszk Seredynski : (see LA)

Abstract:
Keywords:
Roman Slowinski, Jerzy Stefanowski : (see LA)
Using expert's knowledge in rule-based classification of objects
Abstract: The problem being addressed is the classification of objects using decision rules learned from examples. The classification system is taking into account the elements of the expert's knowledge to solve cases where the classified object matches condition parts of few decision rules suggesting different decisions or where it does not match any rule. The support is based on the use of the valued closeness relation.
Keywords: Classification of objects, decision support systems, machine learning, decision rules.
Andrzej Skowron : (see LA)

Abstract:
Keywords:
Jaroslaw Stepaniuk and Marek Kretowski : (see LA)
Decision System Based on Tolerance Rough Sets
Abstract: We present a decision system, which is based on tolerance relations and the rough set theory. It consists from a few almost separate subsystems: searching for tolerance thresholds, data reduction, tolerance decision rules' generation. We investigate various similarity measures and tolerance thresholds to find out tolerance sets. We propose technique whose aim is to reduce the number of examples and the number of attributes involved in the process of learning from examples. We also present algorithm for decision rules' generation and classification of new instances.
Keywords: rough sets, machine learning, knowledge discovery in databases, genetic algorithms
Zbigniew Swiatnicki Roman Wantoch-Rekowski : (see LA)
ZASTOSOWANIE SIECI NEURONOWEJ W EKSPERCKIM SYSTEMIE DIAGNOSTYKI MEDYCZNEJ
Abstract: W pracy przedstawiono ekspercki system diagnostyki medycznej dla potrzeb ortodoncji. System jest eksploatowany w Zakˆadzie Ortodoncji Akademii Medycznej w Warszawie. Om¢wiono jego struktur© i zasady dziaˆania. Opisano problem analizy zdj©† telerentgenowskich wykorzystywanych w systemie. Przeanalizowano mo¾liwo˜ci zastosowania sieci neuronowych do rozpoznawania obraz¢w w eksperckim systemie diagnostyki medycznej. Przedstawiono wnioski z wykorzystania sieci neuronowej jako elementu hybrydowego systemu eksperckiego.
Keywords:
Wieslaw Szczesny : (see LA)
ANALYSIS OF EFFECTIVENESS MEASURES OF BUSINESS UNITS IN A LARGE BANK
Abstract: One of the main problems in a large corporation is the need to compare the effectiveness of particular business units which form that corporation. A business unit may mean a branch (an agency, sub-office, etc.) or even its separate section, for example one that deals with a certain group of clients or offers services of a certain type. In order to judge and to compare the effectiveness of the units in question, when treated as a whole, one usually needs to decide on a set of indices. Each one of them is used to look at a business unit from a different point of view (i.e. it helps to appraise a different aspect of the unit's performance).

Business units do not constitute a homogeneous group, because of variations in local conditions. For that reason, it is difficult to judge which units perform best and which criteria are strongly dependent on one another. It is also difficult to divide business units into groups of similar nature.

The approach described in this paper is based on statistical methods established by the Department of Statistical Data Analysis of the Institute of Computer Science PAS. The methods refer to correspondence analysis and to posterior creation of clusters both of business units and of indices. The methods suggested are applied to the data (appropriately secured) from the bank where the author of this paper works. Two data sets are considered: earnings of bank branches and effectiveness indices of bank branches.
Keywords: statistical data analysis, cluster analysis, stochastic dependence, latent trait, hidden scheme, visualisation, standardization

Wieslaw Traczyk : (see LA)
Qualitative Rules with Quantitative Operations
Abstract: This paper describes two approaches simplifying utilization of qualitative rules with linguistic variables. Assuming ordered sets of terms the first method calculates values in conclusion using special functions, with values from premises taken as arguments. To make these functions computationally manageable, the terms are transformed into numbers, then the function is computed, and result is converted into term. The second method uses numbers instead of terms as well, but the proper conclusion is calculated on the basis of cutting planes, separating subspaces relevant to the different values of conclusion.

Presented approaches reduce number of rules and time of matching. They help to generalize rules preserving initial semantic and therefore such rules can be used for classification of the new cases, unseen during the learning process.
Keywords: qualitative modeling, rule-based systems, machine learning

Jakub Wroblewski : (see LA)
Genetic approach to scalar quantisation
Abstract: This paper considers new techniques supporting the classical centroid algorithm in the scalar optimal quantisation problem: genetic algorithm based on the ``promising object'' method. Experimental results on the grey-scale image data are presented.
Keywords:
Wojtek Ziarko : (see LA)

Abstract:
Keywords:
Jan M. Zytkow and Robert Zembowicz : (see LA)
Discovering Patterns at Different Scale in Massive Data
Abstract: An important but neglected aspect of data analysis is discovering phenomena at different scale in the same data. Scale, also called tolerance, plays the role analogous to error, which is used to focus data exploration on details that exceed error and to disregard details smaller than error. We introduce a discovery mechanism that applies to bi-variate patterns, including time series. It seeks patterns at the full range of scale levels. If it cannot find a regularity for all data, it uses patterns discovered so far to divide data into subsets, and explores recursively each data subset. This search, applied at many scales and to many data sets, seems explosive, but it terminates surprisingly fast because of data reduction and the requirements of pattern stability and significance. We show application of our method on a time series of a half million datapoints. Our example shows that simple data can reveal many surprising phenomena, leading to fine conclusions about the environment in which the data have been gathered.
Keywords: scale, tolerance, discovery, regularities, phenomena at different scale.
xxxxxxxxxxxxxxxx : (see LA)

Abstract:
Keywords:
Thank you for your interest in this Workshop.

This page is maintained by M.A.Klopotek .

If you are lecturer at WIS'95, please deliver your paper no later than till August 15th.
e-mail: <klopotek@ipipan.waw.pl>

Last update: 6th September 1995