CoCo: coding cost for parameter-free outlier detection
Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining
Research on modular-product's cost control and decision-making system based on CBR
CCDC'09 Proceedings of the 21st annual international conference on Chinese Control and Decision Conference
Detecting difference of usage of terms as difference of structure
Cognitive Systems Research
| Hi-index | 0.00 |
When class distribution changes, some pieces of knowledgepreviously acquired become worthless, and the existenceof such knowledge may hinder acquisition of newknowledge. This paper proposes an adaptive Ripple DownRules (RDR) method based on the Minimum DescriptionLength Principle aiming at knowledge acquisition in a dynamicallychanging enviromnent. To cope with the changeof class distribution, knowledge deletion is carried out aswell as knowledge acquisition so that useless knowledge isproperly discarded. To cope with the change of the sourceof knowledge, RDR knowledge based systems can be constructedadaptively by acquiring knowledge from both domainexperts and data. By incorporating inductive learningmethods, knowledge acquision can be carried out evenwhen only either data or experts are available by switchingthe source of knowledge from domain experts to dataand vice versa at any time of knowledge acquisition. Sinceexperts need not be available all the time, it contributes toreducing the cost of personnel expenses. Experiments wereconducted by simulating the change of the source of knowledgeand the change of class distribution using the datasetsin UCI repository. The results are encouraging.