 Dimensionality reduction techniques have been shown to simplify complex hand kinematics by reducing the number of degrees of freedom required to control a high-dimensional hand. Training practices that make the relationship between the low-dimensional controls and the high-dimensional system more apparent can improve learning of an autoencoder-based controller. Three studies are presented which explore different factors that contribute to learning difficulty when using a low-dimensional controller to control a high-dimensional hand. Computer mouse and myoelectric control are compared as one factor contributing to learning difficulty. Training paradigms are explored which include full-dimensional tasks, implicit 2D tasks and explicit 2D tasks. It is found that myoelectric control does not pose a large challenge to learning the low-dimensional controller and is not the primary cause of poor performance. Implicit 2D tasks are found to be as effective as direct training on the high-dimensional hand, while explicit 2D tasks are found to be the most beneficial for improving learning. Establishing an explicit connection between the low-dimensional control space and the high-dimensional hand movements is key to successful learning of the controller. This article was authored by Alexandra A. Portnovariva, Fabio Rosolio, Mora Casadio and others. We are article.tv, links in the description below.