Esprit Proposal No 20288 Acronym: CRIT-2
Proposal Title Cooperative Research in Information Technology
Partner No. 5 Institute of Computer Science of the Polish Academy of Sciences -ICS
Country Poland
Workpackage # 3
Workpackage Acronym : ICS-FAI
Title: Elaboration of Standard Internet Tools for Integration of Databases, Knowledge Bases and Reasoning Systems
Coordinator Maciej Michalewicz

DETAILED DESCRIPTION OF THE RESEARCH PROPOSAL

Elaboration of Standard Internet Tools for Integration of Databases, Knowledge Bases and Reasoning Systems

E.1. Description of the ongoing project the proposal is based on

The proposal described below is based on 2 ongoing grants:

PROJECT 1: Consultation System in the Domain of Hand Wound Surgery

The goal of the project realized under the grant No 1183/94/IS entitled "Consultation System in the Domain of Hand Wound Surgery" financed by the Foundation of Polish-German Cooperation (Sfd-pZ - Stiftung fuer deutsch-polnische Zusammenarbeit / FWPN - Fundacja Współpracy Polsko-Niemieckiej) was to create a computer consultation system in the domain of hand wound surgery.

The project is running since October 1st, 1994 till February 28th, 1997, including reporting and reviewing process.

E.1.1.P1. Present state of knowledge

Hand surgery is one of the most important problems of surgery,both for diagnosis and therapy.A diagnostic system in the area of hand surgery should not only propose a single diagnostic decision (as it would not suffice to launch complex therapeutic activities) but a whole strategy of diagnostic steps concerning the results of individual subgroups of investigations.

E.1.2.P1. Objectives

The system elaborated within the framework of the grant constitutes a foundation for creation of an encyclopedia of knowledge on diagnosis and therapy of hand surgery addressed to every physician. The system has been constructed as a consultation system (so-called expert system).

The system is bilingual: Polish and German.

The system fulfills the following functions:

E.1.3.P1. Significance of the proposed research and expected achievements

The realization of the project should move the cooperation between the Roland Klinik in Bremen/Germany and the Scalds and Plastic Surgery Ward of the Central Clinical Hospital of Military Academy of Medicine/Warsaw onto a qualitatively new stage, as the system gives: the possibility to compare the results in both centers, the possibility of direct exchange of documentation on individual cases, a significant extension of experiences in connection with different profile of hand wounds and their different geography in Germany and Poland, unification of documentation, that creates the possibility of joint scientific publications support for mutual testing of new methods, drugs and apparatus.

E.1.4.P1. Technical description

The system comprises a data base module, a knowledge-from-data extraction module, a bayesian expert system, a deterministic educational expert system and a dictionary module.The heart of the system is a deterministic educational expert system, consisting of a deterministic expert system shell and a hypertext knowledge base.The expert system shell communicates with the database system and a HTTPD server to receive data from the user and to run elementary deduction steps both providing the user with some conclusions and making on-the-fly decisions how to present the data. All the system modules are implemented as CGI-programs started by a HTTPD-server under MS Windows NT or 3.1. A special knowledge description language consisting of instructions of the standard HTML-Language (Hypertext Markup Language) and of special instructions controlling evaluation of the data entered and the selection of the diagnostic path.

E.1.5.P1. Organization and Management

The project is situated in the Central Clinical Hospital of Military Academy of Medicine/Warsaw. It is directed by Prof. Dr. habil. Jerzy Strużyna, the head of the Scalds and Plastic Surgery Ward.

PROJECT 2: Knowledge Discovery in Distributed Databases for Intelligent Query Answering

The goal of the project entitled "Knowledge Discovery in Distributed Databases for Intelligent Query Answering" realized in the grant No 8 T11C 027 10 financed by the State Committee on Scientific Research is integration of several research methodologies created independently in three different research centers, an extended model of intelligent distributed intelligent system is to be proposed. The project is running since January 1st, 1996 till December 31st, 1997, including reporting and reviewing process.

In this project, a system for discovery of deterministic knowledge, based on experiences of the 49-er system, will present its results in terms of equations or taxonomies. Non-deterministic knowledge will be discovered in form of a bayesian network. The integrative task concentrates around the extension of an Cooperative Query Answering system, that has used only knowledge in terms of rules so far, to make use of knowledge in form of equations, taxonomies and bayesian networks during the process of locally unresolved queries. Elaboration of proper techniques of knowledge transformation between the listed formalisms of knowledge representation.

E.1.1.P2. Present state of knowledge

The research group presenting this proposal has been working for many years on the problems of knowledge representation and representation transformations. In these area there exists for five years now close cooperation between the Institute of Computer Science of polish Academy of Sciences (ICS PAS) and the Institute of Computer Sciences of Warsaw University of Technology as well as between the AI Foundations Group of ICS PAS and the Department of Computer Science of the University of North Carolina in Charlotte and the Machine Discovery Laboratory of the Computer Science Faculty of the Wichita State University in Kansas.

In empirical research, after collecting data and introductory exploration of data the researcher put frequently a hypothesis the checking of which is frequently impossible due to missing values of some attributes. Completing the database from "nature" or from other databases or change of hypothesis formulation by the researcher may prove impossible. However, frequently related data bases exist - samples from the same population, elaborated for different purposes, and possessing partially overlapping set of attributes and containing the missing attributes. Such situations are frequent in data bases collected by different medical centers.

E.1.2.P2. Objectives

The system realized within the grant presents methodology of answering unresolvable queries by an intelligent distributed information system. An unreachable query should be understood as the one query for which there exists at least one non-local value of an attribute for a given node. Application of knowledge discovery systems in databases for intelligent query answering is particularly justified as the discovered knowledge can be used to create computational procedures used within the framework of a query to supply unknown values of an attribute. Knowledge discovery in databases can be achieved using various methods. Within this project two methods are to be developed in parallel: (1) bayesian networks and (2) 49er. In this way advantages of both methodologies will be exploited and interesting comparative material will be collected.

E.1.3.P2. Significance of the proposed research and expected achievements

The realization of the project opens new perspectives for cooperation of systems working so far in separated domains like knowledge discovery, probabilistic reasoning, query answering. The project offers common base enabling to communicate between systems from the domains mentioned, showing how results of one system may be exploited in another system.

E.1.4.P2. Technical description

Each node of the system is equipped with a knowledge system (relational database and knowledge base), a system for communication among the nodes and a system for queries answering. The knowledge base contains equations representing numerical dependencies among the attributes used in the database as well as rules and a causal network describing semantic relationships among the attributes. In the project a number of theoretical problems is solved. First of all a new algorithms and tools for the generation of the causal network are proposed. Next a methodology for converting the network into a set of production rules is given. Finally some problems concerning logical integrity of the knowledge base are discussed and an extension of the query language (allowing numerical queries) is proposed. Causal network is nothing but a graphical knowledge base. It can be generated with the help of deterministic algorithms (producing trees and their extension - polytrees) or with the help of probabilistic algorithms, particularly - genetic algorithms. This last approach allows to produce "optimal" (in a sense prespecified by a user) graphical structures.

Having the network and using implemented algorithms of local computations we are able to solve problems belonging to three categories: (a) finding a most plausible value of an attribute knowing the values of a group (specified by the user) of attributes, (b) query answering, and (c) finding most probable explanation of a given observation. Another problem is answering unresolved queries. To solve it, a node sends the query to the remaining nodes of the distributed system. Answers to this query are send by appropriate nodes in the optimal form. We attempt to extend this idea by the production rules obtained from the causal network as well as by the numerical equations generated by the 49er system. This way a portion of knowledge expressed in a mixed form (rules+equations) allows to replace the unreachable values of attributes by the locally reachable terms. Since each node of the distributed system can be queried simultaneously by many other nodes we must solve the problem of an effective communication among the nodes of the system. This problem is supposedly solved by using genetic algorithms.

E.1.5.P2. Organization and Management

The project is placed in the Institute of Computer Science, Polish Academy of Sciences. It is the second scientific project granted from KBN and chaired by dr. Maciej Michalewicz. The first project realized in the period 1991-1994 was concerned with the integration of databases and knowledge bases. At present we suggest an extension of the both projects towards elaboration of standard tools for the integration of databases, knowledge bases and inference systems with the help of Internet.

PROJECT CRIT-2: Elaboration of Standard Internet Tools for Integration of Databases, Knowledge Bases and Reasoning Systems

E.2. CRIT 2 Extension

In summary, the consultation system in the domain of hand wound surgery proposed a new methodology for medical systems, particularly: using standard Internet tools we are able to separate the place where medical test are performed from the place where the full system is located, and we merge different reasoning techniques under common platform.On the other hand, the second system concerning query answering in a distributed information system offers: easy access to the "outer" knowledge by querying different nodes, and integration of different methodologies of data mining by activating specialized tool and translating its output to the common form.

E.2.1. Objectives

The experience gained during realization of both projects shows the need of extending functionality of medical systems by extracting the next modules:
  1. Knowledge transmission among the tools for data analysis,
  2. Data transmission among the input/output of the data analysis modules,
  3. Tools for knowledge generation from data and its visualization,
  4. Tools for reasoning with the help of the knowledge just generated, for explanation of the results and for queries analysis,
  5. Tools for decision analysis,
  6. Multilanguage data and results presentation,
  7. Managing multilanguage databases.
Further there is a real need for a general and standard interface which allow to add, by Internet, different tools for data analysis during a local analysis of a concrete database. As the data can be processed by "outer" systems they must be presented in a "censored" form; designing some safety standards is another problem which must be solved in such a distributed system.

E.2.2. Significance

To obtain better flexibility and modularization of the system for distributed queries the already mentioned functional blocks must be designed. In case of the consultation system in the domain of hand wound surgery this solution allows for future extensions of the set of diagnostic-prognostic tools. By communication with the use of Internet the data analysis project becomes practically independent of the hardware-software platform. Assuming migration of data and not the programs we simplify the process of software development (since it may be oriented toward any platform). Further, decision analysis tools will help the user in planing medical tests and in performing health care activity.

E.2.3. Technical description

Communication with the system is realized via Internet. Software will take the form of CGI programs cooperating with WWW servers and the form of applets (produced under JAVA language) on HTML pages. The tools for filling null values in data will be realized by using the local computation technique in causal networks. By parameterizing the algorithm we will be able to solve (in a unified framework) problems of different nature, like: reasoning, explanation, decision analysis, and queries analysis. To extract knowledge from data genetic algorithms will be employed. This guarantees concise knowledge representation. Particularly, the rules produced from such a representation will have minimal number of prerequisites what makes them readable for a potential user. To attain multilanguage access to the system a technique of vocabularies with extended HTML language equipped with conditional (context sensitive) text generation will be used.

E.2.4. Organization and Management

The main part of the project is unchanged. The systems must be equipped only with standard interfaces to the tools. This way different tools working on different platforms will work under common - internet - ground.

Thank you for your interest in this Project

This page is maintained by M.A.Klopotek .
Please e-mail any comments on this page to
e-mail: <klopotek@ipipan.waw.pl>
Direct access to this home page is via the following http address:

http://www.ipipan.waw.pl/~klopotek/crit2/
Last update: 26th May, 1998 / Created: 20th May, 1998
[BACK] [BEGINNING]