When we solve a problem, we initially have no knowledge and we memorize the raw data. Finally we have general knowledge for solving the problem. To simulate this learning process, we propose a learning method by switching different levels of knowledge representations, each has a knowledge-set, a knowledge generation method and several reasoning methods. The system reasons the class of given data and the successive incorrect classifications trigger to learn with switching the reasoning method to another, reconstructing the knowledge-set, or switching the knowledge representation to the suitable one for the given data. We implement a prototype of the system and apply it to Pen-Based Recognition of Handwritten Digits in UCI Machine Learning Repository. In the simulation, the system can switch suitable knowledge representations from the raw data to general knowledge such as fuzzy rules with a good rate correctly classified and a small amount of knowledge.
View full abstract