Active Transfer Learning for Latent Competencies

Training systems and programs can be made more efficient and effective by understanding how knowledge of one domain affects a learner's ability to acquire skills in another. This Phase I STTR will result in a novel method for modeling this transfer process and predicting when transfer takes place. Underlying this method is a machine learning algorithm that actively solicits input from Subject Matter Experts (SMEs) to learn the latent competencies that influence the ability of learners to transfer knowledge across subjects. The outputs of the algorithm include domain models that can be used in a variety of intelligent tutoring systems (ITS) and algorithms that combine general models with individual views to optimize predictions. The inputs consist of assessments or evaluations of previous learners. Once the machine learning algorithms have learned the structure of one domain, they can more efficiently learn the structure of other domains. In Phase I we will implement the underlying algorithms and create a proof-of-concept software demo that enables SMEs to interact with them. The demo will processes data from two domains and produce results that an ITS will use to adjust how it presents content in the second domain based on learner outcomes in the first.