The National Science Foundation has awarded a three-year, $1.1 million grant to a team led by Yao Xie, Harold R. and Mary Anne Nash Early Career Professor and associate professor in the H. Milton Stewart School of Industrial and Systems Engineering (ISyE). Xie, who is also the associate director for machine learning and data science in Georgia Tech’s Center for Machine Learning, will study “Bridging Statistical Hypothesis Tests and Deep Learning for Reliability and Computational Efficiency.”

“There have been enormous changes in the fields of machine learning and artificial intelligence, with deep learning algorithms developed that have been quite disruptive,” Xie said. “But in theoretical terms, we don’t understand very well how and why these algorithms, which use complex neural networks, work the way they do.”

Her project aims to build a bridge between statistical hypothesis testing and modern machine learning, leveraging deep learning to develop efficient, powerful testing tools for high-dimensional and complex data (akin to the role hypothesis testing has played in previous decades), and also use-testing to develop principled validation tools for machine learning models and provide the foundation of deep models themselves. It’s essential to push the theoretical understanding of deep learning algorithms, so they have what Xie calls “proof of reliability”; in other words, users can know whether or not these algorithms will be reliable in practical situations, such as with stochastic power systems or supply chain issues. Other end-use cases include disease outbreak detection and healthcare systems.

The research tasks are built on multidisciplinary expertise and strong collaborations between the project’s co-PIs: These include ISyE faculty members George Lan, A. Russell Chandler III Professor, and Tuo Zhao, assistant professor; Mark Davenport, associate professor in the School of Electrical and Computer Engineering; and Xiuyuan Cheng, assistant professor of mathematics at Duke University.