A Framework for Model-Based Adaptive Training

Generality: How broadly the model can be applied

A fundamental property of a domain model is its generality or range of applicability. Different levels of generality in the way a model is operating can affect efficiency. Specific reasoning methods are very efficient, but can typically only be applied in few situations. Models that are very general are more flexible for a variety of tasks. A categorisation of levels of generality is needed to establish a way to systematically select the best model for a given training objective.

In the MOBAT framework, the generality or range of applicability is determined by the problem solving operation which is classified in terms of three methods: general, coordinated or fixed. These methods can be supported by:

  1. a principled model which is the most general;
  2. an associative model which is less general; or
  3. a procedural model which is the least general.

Each problem space entity can be realised with different models. Figure 6-5 Partial Workmanship-MOBAT Problem Space contains a sample procedural, associative and principled model. These models are implemented as expert systems (including an explanation facility). The generality of the qualitative models is further specified with a variety of explicit problem solving methods using the RIME methodology, e.g., with the propose, evaluate, reject, apply and elaborate knowledge roles [van de Brug et al. 1986]. The use of knowledge roles provide guidance and flexibility to compose problem solving methods which explicitly specify the differences in the way a model is operating.

A domain model is a subject specific component in the training system architecture therefore subject specific tasks and methods may need to be represented. Figure 6-6 shows the Workmanship-MOBAT representation for tasks and methods. For a given training goal, the first task “to-interpret-topic” points to a particular printed circuit board example (case-model) and results in the application of default generic methods using the propose and apply knowledge roles in “task-planning” and “task-execution” steps respectively. The propose step matches observations to possible topics and the apply step extends the case model with a link to a selected topic. The second approach for this task is using a method which is more specific for the task at hand. A set of steps to accomplish the task starts with:

  1. the identification of keywords in the case model observations,
  2. each keyword is matched with text found in the document chapters,
  3. the possible relevant chapters are listed and ranked,
  4. the most relevant chapter is selected and
  5. the chosen chapter details are obtained for interpretation by the expert.

This representation shows different levels of generality (i.e., different methods) for accomplishing the same task

Figure 6-6 Generic and Specific Method Representation

Figure 6-6 Generic and Specific Method Representation

As presented with Proto-MOBAT and Scheduling-MOBAT, the primary choice of model type for a principled, associative or procedural model is determined by mapping from the expected trainee goal expertise level and preferred learning mode. For a given problem space, advanced and novice problem solving typically proceeds with different approaches in problem solving. Not only are there differences in the way experts solve problems, but advanced trainees are expected to solve problems faster then novice trainees. The use of explicit problem solving methods for different models provide a clear approach to specify the way a model can operate. These methods are of interest because they provide an insight into the development of expertise and offer techniques for training based on multiple ways to solve a problem.

Flower Show
© | | Sitemap