Wei Hu receives Google Research Scholar award for research on deep learning theory for real-world data

Wei Hu, assistant professor of computer science and engineering, has received a Google Research Scholar award in support of his work on deep learning theory. The Google Research Scholar Program provides support to professors performing innovative, relevant research in computer science and related fields. Through his project, “Deep Learning Theory Under Realistic and Measurable High-Dimensional Data Properties,” Hu aims to formulate theories for deep learning models that take into account the complexity and high-dimensional properties of real-world data.

Deep learning theory is an emerging research area that seeks to develop theoretical frameworks to accurately interpret and explain deep learning models, which have traditionally been operated through experience and trial-and-error. In terms of data distribution, however, existing theoretical analyses fall into one of two camps: they either make no structural assumptions about the data, or they make very specific assumptions that are not reflected in real-world data. The result is that existing theories do not capture the complexities of real-world high-dimensional data, contributing to a persistent disconnect between deep learning theory and practice.

Hu’s goal is to fill this gap by identifying realistic and measurable properties of high-dimensional data and developing deep learning theories that take these properties into account. This means building in data assumptions that are simultaneously flexible enough to accommodate complex real-world datasets, and specific enough to capture significant structural properties and enable more accurate predictions. By finding this balance and building more nuanced structural data assumptions into his theories, Hu aims to advance the field’s understanding of deep learning overall.