Machine learning of inductive bias
Machine learning of inductive bias
Artificial intelligence
Hi-index | 0.00 |
Inductive learning is a valuable tool for knowledge acquisition. We present a new, two-phase algorithm, Concept Agglomeration and Division of Attribute Space (CADIA), to overcome the drawbacks of conventional inductive approaches. Use of background knowledge is made by linking the attributes in a semantic net to model attributes being non-applicable to certain examples or taking on default values. This, together with the ability to generate rules of exceptions makes CADIA a powerful tool. We present concepts learned by CADIA in a subdomain of CAM/CAT, planning automatic tests for printed circuit boards, and show their relevance to knowledge engineering. Results from CADIA can give important hints at poorly structured regions of domain knowledge which have to be revised by experts. For future research, we recommend comparative studies about a “bias towards knowledge engineering”.