Matthew Donald Sweitzer

A new era of message processing research will emerge from the convergence of powerful machine learning algorithms with dynamic data from everyday devices equipped with biological sensors. Our study takes critical steps into this era by integrating theory-guided artificial neural networks with eye movements to understand how people learn science concepts from dynamic multimedia. Essential to our theory-guided machine learning approach is a cognitive conceptualization of time as the dynamic interdependence between past and new information that guides how multimedia is attended to and understood. We tracked the eye movements of 197 children as they watched an educational video. We trained two neural network architectures differing in theory guidance to predict learning outcomes using eye movements. The theory-guided architecture, which considered the temporal interdependence of information, yielded more accurate out-of-sample predictions. Our work advances the use of theory-guided machine learning and the development of systems that monitor real-time learning.

We study the preservation of topological characteristics within the temporal link prediction architectures while proposing a model of our own.

Goes over a simple software library (Python) for utilizing sparse autoencoders for more models than just language models.