Assumptions relaxation in Machine Learning Applied to Education

Education is fundamental to raise people out of poverty, improving their standard of living, and creating prosperous and stable societies. Educational technologies  are helping improve access to education (e.g., free Massively Open Online Courses (MOOCs), Wikipedia, etc.) as well as support student success  (e.g., active learning classroom and intelligent tutoring systems). In particular, advancements in educational technologies have strongly relied on the application of machine learning and its many sub-fields. For instance, the research field of adaptive testing aims to improve student competency evaluation while reducing the number of questions being asked. However, in the current state-of-the-art applications of machine learning to education, the models' assumptions about the machine learner fail to transfer to the human learner. Example of such failure include the assumption that the human teacher can design specific sets of questions to fit the constraints imposed on the curriculum by an algorithm.

The goal of this thesis is to address this failure by proposing a model of adaptive testing that does not require the human teachers to design a constrained set of test questions. However, to address this failure, we first must address the common assumption that human learners are adequately equipped to access educational tools (i.e., "Everyone has a device"). Educational technologies are rarely evaluated in terms of their ability to reach or be reached by their intended target users. In this work, we want to take a holistic approach in evaluating our adaptive testing system, by first considering how it could be accessed on campus, and then, by evaluating its intrinsic performance. Unlike prior works that assumed students were always adequately equipped, this is the first work to actually measure usage and access to computing resources from operational data (e.g., Wi-Fi, etc.).

Major Advisor: Jennifer Parham-Morcello
Committee: Robin Pappas
Committee: Margaret Burnett
Committee: Glencora Borradaile
Committee: Prasad Tadepalli
GCR: Brett Tyler