TARGET: Transfer via Active Requests to Generalize Effective Training

Active Transfer Learning (ATL) is a machine learning approach that produces excellent accuracy and predictive power while requiring much less input data than competing approaches. SoarTech, partnered with University of California Davis, has used ATL to improve the understanding of skills and skill relationships in the Navys tactics and decision-making assessment system, DARTS. SoarTech showed in Phase I that ATL let DARTS accurately estimate mastery of sixteen different skills after inputs of only five student data points. We implemented a working prototype and evaluated it with a series of historical and simulated datasets.During Phase II, SoarTech and partners will extend the Phase I research by enhancing the integration of ATL with the DARTS system, addressing limitations in the state of the art specific to understanding complex operational skills, and using ATL to enable new capabilities in training and personnel management such as crowdsourcing to capture new knowledge and updates from operational users. We will evaluate and validate the usefulness of the ATL approach in a series of studies using simulated students and human users in the Option period. The research will lead to new, more detailed, and more frequently updated understanding of skills for training and personnel management experts.