Previous: Propose & revise Up: A task independent description of role limiting methods Next: Comparison between three generic models
This section describes the original `extrapolate from a similar case' approach [Offutt, 1988][Marcus, 1988]. The knowledge acquisition system SIZZLE [Offutt, 1988] was created specifically for sizing problems; tasks which have the goal to find the minimum acceptable capacities a system must have, given the use of the system. An example of a szing task is the identification of minimum computer hardware required to support an office with five secretaries and twenty accountants. The `extrapolate from a similar case' ( ESC) approach is not necessarily restricted to the sizing task, but can be used for any task for which the domain assumptions hold. The description we present here will be independent from sizing problems; but we will show how ESC can be applied on sizing tasks and other tasks. We have not extended this description so far that it will cover the complete case based reasoning approaches [Aamodt, 1991][Schank et al., 1986]; because only quantitative extrapolations are handled by the approach described here.
ESC is a RLM in which learning and problem solving are more integrated than in C&D or P&R. Updating the domain models in P&R and C&D is supported by special shells (e.g. MOLE-KA). In ESC the updating of the domain model is another mode of the problem solver; in which the user offers a case together with a solution instead of providing just the case. In the task decomposition and inference structure we have provided we are for the moment only showing an analysis of the problem solving mode of the the ESC approach.
___________________________________________________