A Framework for Model-Based Adaptive Training

Model Dimensions

One novel part of research in ITSIE is the analysis of subject domain models along different modelling dimensions. This part of modelling should not be confused with KADS knowledge engineering models which are intended to guide knowledge engineering within the life-cycle process [Tansley & Hayball 1993].

The ITSIE approach to multiple modelling dimensions was originally developed for domain modelling based on physical systems [ITSIE 1992, Leitch, Wiegand & Quek 1990] and has not yet been explored in areas which do not have a clearly identified physical system. A physical system domain limits the complexity of the environment; it provides a state where the effects of all actions are know, or accurately predictable [Rasmussen 1986]. The theory for multiple models in ITSIE involves a set of seven model properties:

  1. ontology – the source knowledge;
  2. scope – the physical extent;
  3. generality – the range of applicability;
  4. utility or perspicuity – the ease of use;
  5. abstraction – the precision of description;
  6. approximation – accuracy of description; and,
  7. uncertainty – the belief in the description [ITSIE 1992, D7].

Crafting models that can be used by computers is difficult; the complete ITSIE model characterisation was not validated in applications. The ITSIE project focused on the properties of generality, abstraction and approximation. The MOBIT methodology is essentially similar to ITSIE although the seven modelling dimensions are not explored. In MOBIT, all model aspects are kept constant with the exception of the model generality dimension which is specified in terms of procedural, associative or principled models.

In a tutoring system called QUEST [White & Frederiksen 1990], the problem of model switching is avoided by using a predefined sequence of problems that guide a trainee through a fixed series of domain models. Just one of the ITSIE model characterisations – different levels of approximation – is examined in a system called SAM [Weld 1990]. Reasoning in SAM is in three phases:

  1. model selection;
  2. analysis using the selected model; and,
  3. validation that the assumptions underlying the model were appropriate for the task at hand.

If validation shows the use of an inappropriate domain model then a new model is chosen by an approximation reformulating. Model switching in SAM requires that the model inter-relations are represented.

A single domain model is limited, no matter what the model. For the MOBAT specification framework, the seven ITSIE modelling properties have been extended and linked to properties in the training system. The specification and mapping of modelling properties provide flexibility in creating a training system and the use of these properties results in an adaptive training application.
Flower Show
© | | Sitemap