Towards Model Manipulation for Efficient and Effective Simulation and Instructional Methods

Abstract

Reactive learning environments, microworlds, rely on anchored instruction and scaffolding to achieve constructive learning. Microworlds allow a learner to experiment freely and build on existing knowledge. Anchored instruction presents problems in a realistic setting and to avoid overwhelming detail, scaffolding presents assignments at a level of detail that meets the learner's competence. This is highly motivating as real life problems are immediately solved and details are successively revealed as their complexity becomes manageable. In this context, model manipulation techniques could be employed to systematically manipulate detailed continuous behavior into a simpler counterpart. In many cases, the detailed continuous behavior is abstracted into discontinuities resulting in mixed continuous/discrete, hybrid, models. Model manipulation can also be utilized in a reverse mode by teaching the learner where to look for unmodeled higher-order continuous behaviors to gain insight in physical phenomena that may be difficult to observe directly. If simulation is required in microworlds model reduction is critical to handle otherwise prohibitively detailed behaviors. Efficient hybrid behavior generation algorithms also enable exhaustive model analysis to identify worst-case and even possible undesired model behaviors and feed this back to the learner. Furthermore, hybrid behavior generation allows systems that are dynamically reconfigured (e.g., parts are deleted and created), facilitating interactive modeling and simulation where `what if' scenarios can be quickly designed and studied, e.g., by `rewind' and `fast forward' of behaviors. Overall, to reap these benefits and achieve the overall training and knowledge capitalization goals, effective methodologies need to be developed for automatic model manipulation.