A Framework for Model-Based Adaptive Training

The Framework - Summary

R1-SOAR, RIME, Proto-MOBAT, Scheduling-MOBAT and Workmanship-MOBAT all demonstrate specification and realisation methods evolved from analysis and experimentation with industrial application domains. This chapter steps back and summarises the framework for Model-Based Adaptive Training. Many of the points were not presented earlier, in order to avoid the need for excessive detail in the description of the application research work. This chapter completes the framework with a trainee profile specification. The principled and extensive framework presented in this work consists of a set of methods best suited to implementing a model-based adaptive training system. These methods can be divided into two main areas: problem specification methods (analysis) and system realisation methods (design). In turn, each of these areas subdivide into more specific areas of specification and realisation. These methods are translating general training needs to increasingly specific aspects of the training solution. This bridge from training needs to training solution is structured as follows:

  1. tried and tested analysis methods for a training problem specification which do not consider a computer based realisation (training requirements);
  2. analysis methods for (2a) task decomposition, (2b) expertise classification and (2c) trainee profile and learning style characterisation which consider a model-based computer based realisation;
  3. design methods for training system realisation which separate (3a) what is to be taught (subject expertise) from (3b) the way that it is to be taught (training expertise);
  4. design methods for realisation of expert and trainer training agents which considers multiple modelling dimensions at (4a) the conceptual level and (4b) the implementation level.


This chapter presents the MOBAT specification framework that has resulted from application analysis and experiments. This section starts with an overview of the extended MOBAT framework as shown in Figure 7-1. Each of the main areas in the specification framework is then summarised in separate sections.

It is vital first to characterise the training needs. The framework starts with an analysis of training problems and requirements. This analysis defines those elements that are considered essential for typical industrial training. The initial analysis does not consider computer realisation issues. For example, aspects such as training needs, purpose, topics, target audience and objectives are typical considerations for an industrial training course, whether implemented by computers or delivered by other means. Analysis methods for task decomposition, expertise classification and trainee characterisation are considered separately, as these have a special meaning for the MOBAT framework. The initial training problem specification is decomposed into tasks that the trainee is expected to learn (to perform). For each training objective the associated expertise is classified into goal expertise (i.e., the expected trainee behaviour) and available knowledge in the domain (i.e., the problem space). The trainee profiles are analysed for initial level of knowledge, motivation and ability. The mapping of tasks, expertise and profiles leads to a training analysis for system realisation.

Figure 7-1 The Extended MOBAT Specification Framework

Figure 7-1 The Extended MOBAT Specification Framework

The training system specification defines skeleton training plans and training units (including the most appropriate training strategy). This system specification splits into a detailed design for each of the modules in the MOBAT framework. The design of the user interfaces and specific training tools (e.g., flowcharting and simulation tools) is not considered in detail here. These components can be realised with standard software engineering tools and techniques. The specific emphasis for MOBAT realisation methods are the design specifications of the training agents (i.e., trainer agent and expert agents). The trainer agent design methods are specifications for the required didactic and diagnostic methods. The expert agent design methods are specifications based on the modelling dimensions of the subject models. The required domain models can be based on any combinations of principled models, associative models and procedural models. Both trainer agent and expert agent design methods are split into conceptual design (knowledge level) and system design (symbol level). This specification framework involves quite a large number of factors and subtle issues.

There are several reasons for specifying training at various levels of detail. A major reason for using specification layers is to be able to communicate at different levels to different people. For example, business managers, trainers, experts and trainees all have different views and roles in the training specification process. Business managers are interested in high level issues such as training purpose. Trainers are key contributors to the training analysis, skeleton training plan and trainer design methods. Although trainer authors have a key role, not all specification methods for MOBAT involve trainer authors. For example, domain experts are key contributors to the expert agent realisation for knowledge acquisition. The target audience (i.e., trainees) needs to be examined to define the trainee needs and trainee profiles (which determine issues such as trainee preferences and training strategy).

Another reason for increasing levels of detail is to make adaptability easier; that is, to accommodate change in the training subject matter or change in the training requirements. In an industrial environment where training problems often change, considering adaptability is a key factor to delivering successful training. Increasing levels of detail is also about re-usability of the subject matter. In industry, the training subject is likely to be a specialised (and valuable) knowledge base. Various viewpoints of the subject domain are required, either to teach a particular subject or to re-use the domain subject matter for a different training approach (e.g., on-line reference system). The MOBAT ‘tool set’ approach is supporting both adaptability and re-usability. Pre-developed modules with libraries representing the various options in expertise and instruction can be selected and combined for a particular training application. This requires a common architecture [Slater, Brown, van de Brug & Brown 1984] within which the various modules can be substituted without requiring a re-design and a specification methodology that provides a set of guidelines on how to select and combine the most appropriate modules.

The remaining sections in this chapter are organised as follows. Section 7.3 presents a clear starting point for the MOBAT framework in terms of training purpose and training objectives. Section 7.4 reviews the decomposition of tasks a trainee is expected to learn (to perform). Section 7.5 summarises expertise classification. The trainee characterisation in Section 7.6 is presented in greater detail as this not been presented earlier. Section 7.7 contains a discussion of desirable characteristics for training units. Section 7.8 presents a summary for the system realisation methods. Finally, Section 7.9 contains general conclusions and discussion.
© | | Sitemap