Modeling Training Efficiency in GIFT

摘要

The US Army Learning Model (ALM) emphasizes the importance of deployable, individualized, adaptive training technologies to help Soldiers better learn and improve critical skills in dynamic and challenging environments. The Army is developing one such technology known as the Generalized Intelligent Framework for Tutoring (GIFT). GIFT is an open-source, domain-independent intelligent tutoring framework that facilitates reuse of components in an effort to reduce the expense of developing and delivering adaptive training. Adaptive training offers the promise of higher levels of proficiency, but another important benefit is that it is more efficient than one-size-fits-all training. Put another way, intelligent, adaptive training should require less time to train a population of learners to a given level of proficiency than non-adaptive training. The gains in efficiency should be a function of several factors including learner characteristics (e.g., aptitude, reading ability, prior knowledge), learning methods employed by the adaptive training system, course content (e.g., difficulty and length, adaptability), and test characteristics (e.g., difficulty, number of items). Optimizing training efficiency requires one to tune the instructional design and course content to the characteristics of the learners. GIFT currently lacks the ability to model or predict the efficiency with which training can be delivered based on these factors. This paper presents a process, and proposed architecture to enable GIFT to make estimates of training efficiency. How this architecture supports authoring and how machine learning can be used to improve the predictive model are also discussed.