A Framework for Model-Based Adaptive Training
This section presents a condensed design specification of the scheduling training requirements. The MOBAT system specification is mapping the problem specification (e.g., training need, task decomposition, expertise classification and trainee profile characteristics) to the appropriate learning mode, training strategy, skeleton training plan and training units.
The outcome of training system specification is a separation of what needs to be taught (e.g., the training unit contents with any dependencies) and how to deliver the training (e.g., learning mode, training strategy, training plans). This specification feeds into the MOBAT design work for separate domain expert agent(s), and a generic trainer agent.
Mapping the available order scheduling associative knowledge to the requirement of trainees operating at the level of skill-based expertise suggests that the primary learning mode is deductive. I.e., the knowledge representation model is ‘above’ the desired level of performance.
A skill based performance level and preferred learning mode of deductive can therefore be accomplished with the existing associative model. To serve other expected performance levels and learning preferences, rote and inductive learning modes will also be required. This means knowledge will also need to be represented ‘below’ and at the ‘same’ level of required levels of trainee behaviour.
The Scheduling-MOBAT training strategy is based on the following specification. Scheduling supporting expertise can be presented with a discovery strategy which allows a variety of trainees to navigate the supporting expertise freely as needed. Practicing order scheduling is the main way for the trainee to learn the overall training objective ‘to schedule orders’.
Mapping the training strategy as a straightforward function of assumed average trainee self-supporting level suggests a trainer facilitating strategy to practice order scheduling. For those trainees that need a scheduling demonstration, a coaching strategy is specified to provide a step-by-step expert demonstration.
Should a trainee need to learn a particular scheduling task in detail and a low self-supporting level is apparent, then a tutoring strategy with extensive trainer control is suggested. Depending on trainee performance, alternative strategies can be realised as needed at run time.
A possible skeleton course map for the order scheduling objective is shown in Figure 5-3. This is not a fixed pre-defined plan (course curriculum). The skeleton plan is a possible input from the trainer interface in the training system architecture (Figure 1-1).
Defining the skeleton training plan for Scheduling-MOBAT is done with four general classes of training units. If no skeleton plan has been provided, then the trainer agent will create a plan from the overall training strategy and training goal hierarchy. The emphasis in this work is on the training units, which are dynamically sequenced by the training system rather than on the skeleton training plans.
Training units are self contained fragments of training interaction. To adjust training strategies the trainer needs to be able to measure trainee learning within and across training units. As presented for Proto-MOBAT in Chapter 4, properties of training units have been described which are: training unit type, generic task, problem space, goal expertise, learning mode, training strategy, simulation model, domain model type and supporting expertise.
To measure trainee learning, this section presents two additional properties based on (1) the transfer of learning method and (2) expected time duration of training units.
The training system itself needs to embody principles of learning. The principles of learning in this research work are based on a transfer of learning categorisation as shown in Section 2.6.1 with R1-SOAR experiments. The transfer of learning measurements are classified as within-trial, across-task and across-trial. As shown with R1-SOAR, it is possible to implement an expert system with these transfer of learning methods.
Within-trial learning can be recognised, with small chunks of knowledge as they are re-used to accomplish a task. Across-task learning can be recognised, with chunks of knowledge for a particular situation being transferred from one task to another (i.e., where the same action is appropriate from one task to another). Across-trial learning is generally recognised when a particular task is repeated correctly with a sequence of operations. An indication of measuring learning and development is provided for each of the types of training units below.
Presentation Training Unit: The presentation training unit provides an overview of order scheduling concepts. It is an introduction to order scheduling applicable for all training tasks. The measurable chunks of learning for the presentation training unit are represented as supporting expertise. This training unit consists of a sequence of concepts that the trainee is expected to grasp as new concepts are presented (within the same training unit). The learning of supporting expertise is measurable from trainer and trainee communication. Transfer of learning difficulty may be identified if a trainee reviews or requires a particular concept to be explained. If the trainee needs the same supporting expertise again-and-again (within the same training unit) then the diagnostic tactician may determine this to be a misconception.
Demonstration Training Unit: The demonstration training unit provides the capability to schedule a set of sample orders with the expert controlling the step-by-step scheduling process. An expert demonstration of order scheduling is applicable for all training objectives. The chunks of learning for the demonstration training unit are procedures for sample orders. Across-trial failure or success of learning can be measured in demonstration training units, depending on a trainee’s ability to repeat the order scheduling task as demonstrated by the domain expert.
Practice Training Units: An order scheduling practice training unit provides the trainee with a “learning by doing” approach to scheduling a set of orders. The chunks of learning for practice training units is the knowledge to take appropriate action depending on a situation. The training unit is measurable when a trainee attempts new tasks, and shows the ability to apply these chunks of learning from previous tasks (across-task transfer of learning). If a chunk in the order scheduling practice training unit has not been learned, then the trainee can practice each of the scheduling tasks in separate training units. Chunks of learning in these detailed training units cover small pieces of knowledge that make up the scheduling rules (e.g., “to interpret validate date” training unit in Figure 5-3 Skeleton Training Plan for Order Scheduling Objective). These small chunks of knowledge are often re-used several times while solving the same task therefore this is measurable by recognising within-trial learning methods. Details of design realisation for small chunks of knowledge are presented in Section 5.6.
Assessment Training Unit: These are units of training to formally test trainee performance for each training objective. The assessment training unit covers a sample set of scheduling orders. The chunks of learning for the assessment training unit are represented as associative knowledge. Learning should occur before the formal assessment takes place. With a trial-and-error approach for a particular task, a measurement of across-trial transfer of learning is possible in assessment training units.A design consideration for training units should be an estimate of the time taken by a target trainee to complete a training unit. A trainer or business input on this attribute can affect the realisation level of detail or approach to implementing training units. The trainer agent may wish to adjust strategy, or intervene, depending on the time taken by a trainee to complete a training unit. Detailed time estimates for a different target audience may be appropriate. The training unit time estimate for Scheduling-MOBAT experiments are designed as follows. The average trainee is expected to complete the scheduling presentation in 1.5 hours, demonstration of order scheduling in 30 minutes, practice order scheduling 2.5 hours, detailed task practice (if necessary) of 5 minutes for each scheduling rule and assessment of 5 orders in 20 minutes.
© RuleWorks.co.uk | | Sitemap