Theory Restructuring: Coarse-grained Integration of Strategies for Induction & Maintenance of Knowledge Bases
Edgar Sommer · 1996
Even after the publication of (Michalski & Tecuci 1994), the debate over what precisely constitutes multistrategy learning continues. While some authors have advocated a very fine-grained view of cooperation between relatively small learning operators, (Erode et al. 1993) argued that significant practical benefits can be obtained from a more coarsegrained interaction between different learning modules, each of which is complete by itself. I subscribe to this view, but add non-inductive functionality to establish a multistrategy context for the design and maintenance of knowledge bases. This paper describes (the implementation of) a coarsegrained, multisU’ategy architecture for theory induction and maintenance, or theory, restructuring, called RRT. The aim of theory restructuring, and RRT, is to close the gaps between knowledge engineering, knowledge acquisition and machine learning, in order to better support the design, and maintenance over time, of knowledge bases. To this end, a number of subtasks in the KB-maintenance process have been idcntiffed, namely ¯ induction: offering access to existing implementations of approaches to machine learning in the hope of automatically generating useful rules from exemplary data, ¯ analysis: offering insight into the status quo of a KB.