A Framework for Model-Based Adaptive Training

To adjust the current domain model

Depending on trainee performance, the diagnostic tactician within the trainer agent is used to adjust models when necessary. Diagnosing the cause of incorrect responses from a trainee and taking appropriate action are essential capabilities of an ITS. In Workmanship-MOBAT, the main tasks for the diagnostic tactician are to advise, to intervene, to remediate and to assess. The diagnostic tactician can select and apply a combination of tasks with appropriate methods to adjust a selected domain model. There can be a large number of possible reasons for incorrect trainee behaviour. The diagnostic tactician tasks and methods identify and correct trainee errors in terms of slips, bugs or misconceptions. This classification is similar to the ITSIE types of errors that a trainee can make during a training session. In Workmanship-MOBAT, a propose step is used in the diagnostic problem-space to select the most appropriate trainer task and trainer method combination for identifying and responding to trainee errors. Examples of diagnostic methods are given below.

  • A slip is considered an unstable error which results from a simple trainee mistake. Such slips may normally be remedied by repeating the exercise or the trainer providing some limited advice. For frequent trainee mistakes the trainer may need to apply a slight model adjustment to verify that the error is just a slip and to remind the trainee of correct behaviour. A trainee slip can result in the use of a method (called the “propose-abstraction-method”) to adjust the precision modelling dimension. For example, a presentation steps back to a previous level to remind a trainee of an earlier presented concept, the trainee is given practice at a higher problem level, or a test is given at a slightly higher abstract level.
  • A bug is a clearly categorised and stable trainee error. These are common trainee errors for which the trainer can make selected model adjustments. When the diagnostic tactician detects a trainee bug, the trainer may attempt to correct the error by using the “propose-abstraction-method” as noted above. Alternatively, the trainer may select a method (called the “propose-approximate-method”) to adjust the accuracy modelling dimension. For example, a new presentation range may be appropriate, the trainee is given practice covering a smaller problem range, or the trainee is given a test covering a narrow topic range for a given bug classification.
  • A misconception is an uncommon but stable trainee error. Both bugs and misconceptions are erroneous beliefs by the trainee which may manifest themselves as consistently incorrect behaviour. However, misconceptions are not clearly identified by searching a bug catalogue. There is no formal basis and no recognised solution for the trainer agent to explore potential model changes. This can be considered a blind search in a space of variations [ITSIE 1992, D7]. A trainee misconception may result in the use of a method (called the “propose-commitment-method”) to adjust the uncertainty modelling dimension. For example, an alternative training unit may be presented, practiced or tested by the trainee with a different portrayal for a given problem. It is, of course, not possible to guarantee that changing to an alternative training unit is going to resolve the trainee misconception.

Detailed model-based diagnostic methods have not been investigated in Workmanship-MOBAT. Although there is still a lot of refinement to be done, the explicit modelling dimensions within the domain expert, and the flexibility of using trainer task and trainer method combinations, provide an excellent framework for further development of model-based diagnoses in training application.

Flower Show
© | | Sitemap