ATD: Efficient and Effective Algorithms for Detection of Anomalies in High-dimensional Spatiotemporal Data with Large Amounts of Missing Data

Detecting rare events of anomalies using high-dimensional real-world data with mixed-type multivariate response, sparse and unknown signals and a large number of missing values is very challenging. The architectural integration of domain knowledge and data with spatial and temporal variables significantly improves the model reliability, interpretability, and prediction accuracy. For predicting anomalies, the investigator plans to develop efficient and effective posterior sampling algorithms and apply these methods to various real-word data with interdisciplinary collaborations. Students, including those from underrepresented groups, will be involved in education and outreach activities featuring real world problem-solving. The research aims to develop novel anomaly detection algorithms. These will leverage hybrid information from the domain knowledge (represented as prior distributions) and regularization constraints to establish an efficient and effective search direction. Data information embedded in likelihood functions and regression models will reduce the search space for parameter estimation. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.