Dissertation Proposal - Clayton Fields

Modeling Vision and Language at Human Scales

Presented by Clayton Fields, Data Science emphasis

Hybrid presentation: Attend in-person in City Center Plaza (CCP) 368 or register to attend online via Zoom

Abstract

The impressive results that have recently been achieved in natural language processing have been primarily driven by increasingly large models with many parameters and using enormous, web-scale datasets to train them. The size of models and their training data requirements present costly demands that freeze many researchers out of training with cutting edge models. Furthermore, multi-billion word datasets can generally only be obtained from the internet. Datasets derived from other sources, such as from human beings, are much smaller and have not been extensively studied with current language modeling methods. Current methods also learn from text alone, without the rich array of sensory information that human beings use in learning language. In this dissertation proposal I present a body of research aimed at addressing these concerns.

Committee

Dr. Casey Kennington (Chair), Dr. Edoardo Serra, Dr. Tim Andersen