Optimal static and dynamic training schedules: state models of skill acquisition

Abstract Training takes place in complex environments. Typically, there are many different tasks which need to be learned; each task can be performed at one of several different levels of proficiency and each level of proficiency within a given task can be trained in one of various different ways. Much is known about what tasks need to be trained in order to achieve a particular objective, what methods are best for training a particular level of a particular task, and what measures should be used to evaluate training. Curiously, given that the tasks, methods, and measures have been selected, very little is known about how to determine which level of which task it is best to train in each session. In this article, a framework for pursuing the optimization of training schedules which are sensitive to (dynamic) and not sensitive to (static) the session by session (trial by trial) progress of the learner is developed. The framework takes as its starting point the early state models of learning first proposed within mathematical psychology in the 1950s. We show that the state models can be used to predict how performance will vary as a function of the scheduling of training trials. Practically, it is important to consider the effect of changes in the scheduling of training trials because such changes can substantially reduce the time it takes any given individual to learn a composite skill. Theoretically, it is important to consider the effect of changes in the scheduling of training trials because such changes can potentially provide the answers to a number of questions central to research in training. We conclude that the state models of learning provide both of the hoped for practical and theoretical benefits.