Title: Bayesian Modeling of Self-Exciting Marked Point Processes with Missing Histories
Non-persistent collection of marked point process data in space and time occurs in many remote sensing applications, resulting in known intervals of time where events are unobserved. Self-exciting marked point processes (i.e., Hawkes processes) have been shown to be effective in these situations to understand the underlying process. Recent estimation procedures assume a full history of the data and are not capable of accounting for regions of space and/or time where data are missing. A Bayesian estimation procedure for self-exciting marked point processes is developed to naturally incorporate the missing data mechanism probabilistically. Accounting for the missing data improves estimation and prediction of the process. This is demonstrated through simulation and an application to real conflict monitoring data from the Global Terrorism Database where records over significant time periods are missing.
Bio: John Lewis joined Sandia in 2014 shortly after finishing a Ph.D. in Statistics at The Ohio State University. For his dissertation, he developed methods for conditioning on insufficient statistics in Bayesian models for the purposes of robustness. At Sandia, John supports a variety of projects using a wide range of statistical methodologies. These methodologies include functional data analysis, design, and analysis of physical and computer experiments, inverse prediction, uncertainty quantification, and spatio-temporal modeling.
Contact Name: Li Li
Contact Email: email@example.com