RuleWorks

A Framework for Model-Based Adaptive Training

Uncertainty: The belief in the model description

The uncertainty of a model can be expressed by the belief or confidence in the chosen description of training properties in the subject domain. High confidence in a model description may exist if a model has been used successfully in the past.

Perhaps an alternative model description may be appropriate for different trainees or an adjustment in description is needed if the trainee is not performing as expected with a particular model.

Within ITSIE the uncertainty dimension was not exploited as none of the representations of expertise supported explicit reasoning with uncertainty [ITSIE 1992, D7].

The relative high number of 222 case models in the Workmanship-MOBAT domain provides a useful way of utilising the confidence dimension with two properties.

A case model has been given slots for (1) votes and (2) preferences to explicitly express the confidence in a chosen model. Votes are numerical indicators of confidence. Preferences are based on a ranking scheme using the symbols worst, worse, acceptable, better and best.

The training system fills these slots based on how successful a case model has been used in past training situations. The votes and preferences are represented in case models and are available to the trainer agent in determining the most appropriate model. As in SOAR, the trainer agent has a task selection mechanism which is also based on votes and preferences.

The explicit reasoning with uncertainty is supported with knowledge for proposing and evaluating tasks. If insufficient knowledge is available to make explicit decisions then a fixed decision procedure selects the next model based on highest preference ranking and highest votes.

This mechanism can play a significant role in reducing uncertainty in the training process by selecting the most appropriate model for a given situation.

*
Flower Show
  *  
© RuleWorks.co.uk | | Sitemap