A Framework for Model-Based Adaptive Training

Introduction and Aims of the Research

Model-based training by computers is in the early stages of development. Although convential computer-based training is starting to gain acceptance, there have been few attempts to deploy model-based computer-based training systems in industrial environments. The term model-based here means explicit models of cognitive processes and qualitative models of the physical device, product or process on which the training task is to be performed.

Most existing training systems do not have an explicit model-based approach but are more likely to be based on either a direct representation of the subject matter or a quantitative (numerical) simulation model. Methodological issues for the structure and formulation of model-based computer-based training are hence open and central.

This research is concerned with three methodological issues:

  1. characterising industrial training domains for model-based intelligent training;
  2. mapping elements of the training specification to model-based intelligent training agents;
  3. adaptability to changing industrial training situations.

An arduous task in creating an industrial training system is in specifying all factors in the training domain that can be used effectively by model-based intelligent training agents. In complex training domains a wide variety of aspects are typically relevant at any point of time. Another concern is that decomposing training subject matter at a low level of detail (i.e., decomposing the tasks the trainee is expected to learn at too fine a grain size), may result in complex software structures.

The solution in this research work is to extend the training methodology from [Sime & Leitch 1992] into a framework made up of specification and realisation methods in terms of four core activities:

  1. tried and tested analysis methods for a training problem specification (e.g., see [Gagne, Briggs & Wager 1992]);
  2. analysis methods for task decomposition, expertise classification and trainee characterisation;
  3. design methods for intelligent training system realisation which separate what is to be taught from the way that it is to be taught; and,
  4. design methods for realisation of expert and trainer agents which consider multiple modelling dimensions at a conceptual level and a software implementation level.

The second issue addressed is the mapping of a training problem to the realisation of model based intelligent training agents. The solution in this research work is to specify the relationships between the elements defined by training characterisation in a set of training unit properties, for realisation in a model-based intelligent training architecture (adapted from [Slater, Brown, van de Brug & Brown 1994]). The training architecture is based on a separation of the knowledge that is to be taught, called subject expertise, from the way that it is to be taught, called training expertise. Intelligent training agents are specified at both the knowledge level and symbol level [Newell 1981].

The third issue addressed is adaptability to changing training requirements. Training objectives often change because of new training requirements, changes in the target audience or because various other constraints imposed by the (industrial) environment change. The solution in this research work is to place the elements defined by training characterisation in a general theoretical framework for training and domain expertise specification, that makes it easier to specify new knowledge and evolving contextual details. A specific capability within the framework, promoting adaptability, is in the explicit mapping of a model-based training application with:

  1. a set of problem spaces [van de Brug, Rosenbloom & Newell 1986];
  2. a tool set approach for distinct knowledge representations [MOBIT 1994];
  3. the decomposition of training tasks [Leitch & Gallanti 1992];
  4. the use of explicit problem solving methods [van de Brug, Bachant & McDermott 1985]; and,
  5. the integration of multiple modelling dimensions [ITSIE 1992, D7].
The intended reader of this work is the intelligent tutoring systems community and courseware tool authors interested in putting model-based theories into practice for industrial training environments. The advantages of an intelligent model-based training system (e.g., flexibility, personalised training, effective learning) are widely recognised in the ITS community, but their construction is difficult and time consuming. An adaptive model-based system requires (a) a strong framework showing how to select and apply appropriate models and (b) guidelines for conceptual structures that can be imposed on the knowledge available for industrial training situations. The framework proposed in this work is aimed towards the ultimate goal of delivering Model-based Adaptive Training as a new solution for industrial training problems.
Flower Show
© | | Sitemap