Title: Statistical Learning and Decision Making for Spatio-temporal Data

Date: April 8th, 2021

Time: 11 AM EST

Location: https://bluejeans.com/5007129655

Student Name: Shixiang (Woody) Zhu, Machine Learning Ph.D. Student

Home School: H. Milton Stewart School of Industrial & Systems Engineering
Georgia Institute of Technology

Advisor

Dr. Yao Xie (H. Milton Stewart School of Industrial & Systems Engineering, Georgia Institute of Technology)

Committee

1 Dr. Pascal Van Hentenryck (H. Milton Stewart School of Industrial & Systems Engineering, Georgia Institute of Technology) 

2 Dr. George Nemhauser (H. Milton Stewart School of Industrial & Systems Engineering, Georgia Institute of Technology) 

3 Dr. Feng Qiu (Argonne National Laboratory)

4 Dr. He Wang (H. Milton Stewart School of Industrial & Systems Engineering, Georgia Institute of Technology)

Abstract

Spatio-temporal data modeling and sequential decision analytics are a growing area of research with an enormous amount of modern spatio-temporal data being consistently collected from the real world. These applications include power grids, public safety systems, healthcare systems, financial markets, social media, IoT networks, and even our personal mobile devices. Understanding the intricate spatio-temporal dynamics behind these data requires the next generation of mathematical and statistical algorithms based on quantitative models of human and physical dynamics. In this proposal, we first present the recent developments in this area with both methodological advances and various real-world applications. Then we develop new theoretical and algorithmic techniques for capturing the dynamics of real-world spatio-temporal data by combining cutting-edge machine learning and classical statistical models. We also formulate the sequential decision making process as an optimization problem in a data driven manner, which could suggest better decisions by taking advantage of the historical knowledge. Lastly, we study a wide array of real-world spatio-temporal datasets using our proposed methods. The results demonstrate the value of spatio-temporal analytics in understanding computational, physical, and social systems.