A Framework for Model-Based Adaptive Training

Perspicuity: The ease-of-use of the model representation

The perspicuity of a domain can be considered with respect to particular training situations or examples. It is often possible to represent a domain model with different case-models which enhances the perspicuity but does not effect other properties. Each example or case model has a reduced representation of domain variables (but not a reduced representation of reasoning methods). Case models for the workmanship problem space are defined for sample printed circuit products.

A particular example of printed circuit board which represents only those aspects of the problem to be solved (i.e., a domain specific task) may be easier to use instead of a cluttered printed circuit board which may contain too many distractions for the trainee. In Workmanship-MOBAT, a set of 222 printed circuit board examples are represented as individual case models. Each case model contains attributes and variables for a sample printed circuit board in terms of components and observations.

The trainer agent selects an appropriate case model for a given training objective and compares expert and trainee reasoning to extend the case model with appropriate information. For example, if the training task is to identify the cause of a defect, then the appropriate diagnostic reasoning approach by the expert can be compared with the answers given by the trainee. The case model is extended with information identifying the relevant defect category and possible defect causes.

For a given training task a number of case models may be possible. The trainer agent adopts a relevant training strategy and uses trainer methods to propose an appropriate case model based on perspicuity. A sample case selection heuristic used in the Workmanship-MOBAT system is to prefer the selection of a case model with the least complexity for a novice trainee. I.e., a case model with the most relevant set of domain variables for a given task. For an advanced trainee, if available, a more complex case model with a wide range of domain variables for a given task is preferred.

Figure 6-7 Workmanship Case Model Representation

Figure 6-7 Workmanship Case Model Representation

The Workmanship case model representation is shown in Figure 6-7. Each case model has a unique identification and status attribute. The remaining properties are compound attributes which are lists of pointers to other object structures. These pointers are either generated by the expert agent(s) or tutor agent. The expert agent reasons about the domain model and generates appropriate pointers for each case model (see Figure 6-7 right-hand column).

The tutor agent reasons about a trainee model and generates (dynamically) appropriate training units which are linked to a set of case models properties (see Figure 6-7 left-hand column) as required. The tutor agent also enables a set of expert ‘guides’ which are associated with case models for explanation purposes. The use of ‘guides’ in the training system architecture is described further in [MOBIT 1994, W1, W6].

Flower Show
© | | Sitemap