A Framework for Model-Based Adaptive Training

Scheduling System - Transfer of Learning

The three different types of knowledge presented above make it easier to specify “transfer of learning” methods within the training framework. The purpose of specifying these methods is to define ways of measuring and comparing (with the expert model) the learning of items of expertise (called chunks of knowledge in SOAR) that need to be conveyed to a trainee. In the MOBAT framework, it is proposed to explicitly specify (in training unit properties) the items of expertise together with their transfer of learning method.

The expected type of transfer of learning is based on SOAR’s learning mechanism [Laird et al. 1987]. SOAR uses chunking which is a single learning mechanism for creating new production rules for all aspects of problem solving. Three types of transfer of learning can result from this chunking mechanism:

  1. across-trial transfer of learning;
  2. within-trial transfer of learning; and
  3. across-task transfer of learning (see Table 2-1 R1-SOAR Learning [van de Brug, Rosenbloom & Newell 1986]).

This classification is used in the MOBAT framework to indicate the expected transfer of learning in each training unit. In SOAR the learning chunks are always production rules. For the MOBAT framework the learning chunks (i.e., items of expertise) can be procedures, rules or principles.

Figure 5-9 Transfer of Learning

Figure 5-9 Transfer of Learning

An example of Scheduling-MOBAT transfer of learning chunks is shown in Figure 5-9. Within-trial transfer of learning is typically possible for small chunks of knowledge (e.g., principles). For the training task “to validate date”, the concept in the rules for “today’s date” is used 8 times. This provides the opportunity to re-use a learned chunk of knowledge 7 times for the same task. Across-task transfer of learning is generally applicable for re-use of a learned rule from one task to another. The simplified rule “If Line-Item moved then move members” is used for many different tasks in the scheduling domain. There can be many situations when a line-item is re-scheduled which mean that all the parts belonging to this item must also be re-scheduled. This is a re-usable rule across different tasks. The procedural task “to detect clash” is learned by repeatedly executing the same task. Learning the procedural task in this case is therefore termed across-trial transfer of learning.

Figure 5-9 Transfer of Learning shows examples of rote learning modes. The same transfer of learning categories are also applicable for inductive and deductive learning modes. Learning modes are specified as ‘preferred types of learning’ for individual trainee’s. For a description of learning modes see ITSIE types of learning in Section 3.3.2. The three transfer of learning methods are principles of learning within the training system itself, which fit neatly with the classification of distinct expertise models and preferred learning modes.

© | | Sitemap