Accumulating qualitative knowledge

The author considers the problem of learning qualitative descriptions (landmark points) of dynamic physical systems. Specifically, it is shown that it is possible to accumulate knowledge about landmark points of a dynamic system by invoking machine learning (generalization) rules. The rule proposed here is the rule of `physical similarity´. It is shown how this rule can be used to generalize the knowledge about landmark points. Several learning scenarios are analyzed, and a general reasoning procedure is presented. The algorithms for calculating the similarity modules are implemented in the COPER discovery system