The advancement of point process models in healthcare and security analytics, such as in predicting disease occurrences, crime and cyber security attacks, has the potential to significantly improve public health and safety. Point processes are an important but underappreciated class of models in engineering that incorporate temporal dynamics, information diffusion, and recurrent behavior. This project will establish theoretical foundations and computational methods that enable efficient modeling and estimation of point processes from real-time and large-scale transactional data. To address emerging data-rich challenges in healthcare and security analytics, this project aims to break the modeling and computational limitations of current theory and practice in point processes, providing a powerful new set of tools for these significant engineering challenges. The PIs are committed to devoting their efforts to facilitate the education and training for next-generation engineers and data scientists, especially women and underrepresented minorities.
The development of efficient inferential analysis and decision-making for point process models is far from reaching the same level of maturity as that of Gaussian models. In this project, the PIs will leverage elements from many fields - optimization, machine learning, nonparametric statistics, and information theory, to address several fundamental modeling and computational hurdles in the context of multivariate Hawkes processes. The project has four main research thrusts: (i) investigation of nonparametric models to significantly advance the capability of point processes in capturing large-scale and complex event data; (ii) development of novel theoretically grounded and practically efficient learning and optimization algorithms for statistical inference in point processes; (iii) exploration of domain-specific structure to achieve the optimal trade-off between likelihoods and regularizations; (iv) development of online optimization schemes and tools to facilitate real-time prediction and inference for streaming event data. The techniques developed in this project will be used to advance the modeling and prediction of specific healthcare and security related applications, such as discovery of disease relationships, tracking of adverse drug reactions, and detection of criminal and terrorist activities.
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.
