Ph.D. dissertation: Neda
Title: UNIFORM ALGORITHM FOR INTEGRATING MULTIPLE KNOWLEDGE
BASES INTO ONE
†† We have
designed a uniform algorithm for integrating multiple knowledge bases into one.
The algorithm enables the integration of the arbitrary number of systems and is
domain independent. Integration in general enables better classification
accuracy, and better transparency than any single knowledge base.
Implementation at Rolling Mill Bela of Iron and Steel Works Jesenice has shown
an important increase of about 5 % in terms of classification accuracy. The
system has been regularly used for two years and is one of the major AI industrial
applications in Slovenia.
Contributions of the dissertation:
integrating algorithm is general in the sense of the number of
incorporated single systems and can be applied to most real-life problem
integrating schema enables the encapsulation of expertís knowledge during
the process of integration. The process can be automatic or automated.
- While single
knowledge base can be designed in different ways, the integrating
algorithm extracts more reliable parts and eliminates noise in single
- In addition
to the algorithm for integrating, algorithms for initialisation and
maintenance have been also designed, thus providing a global integrating
methods for testing and comparing single and integrated knowledge bases
have been designed.
- The algorithm
was applied in the integration of four single systems. The implemented
system at Rolling Mill Bela has achieved significantly better
classification accuracy (by 5 %). The system is in regular use and has
shown practical maturity of applied methods.
- The expertís
opinion in that the proposed approach seems appropriate for real-life
practical domains where it is difficult to find one single adequate