Mathematics Colloquia and Seminars

Return to Colloquia & Seminar listing

Compared with the commonly collected health information, long-term and high-frequency physiological time series that exhibit nonstationary traits provide abundant information from an alternative perspective. However, it's challenging to extract clinically applicable data from these time series due to various impediments. Fueled by clinical necessities, physiological expertise, and clinical observations, we introduce a novel latent diffusion geometry-based signal processing method. The algorithm is explicitly designed to handle enormous datasets - including ultra-long and high-frequency time series - with ease, while also remaining resilient to color and heterogeneous noise, underpinned by sound theoretical support. Additionally, we demonstrate that the algorithm effectively reconstructs the spectral structure of the Laplace-Beltrami operator with an L-infinity convergence outcome. We will discuss an application of this method to assess the clinical outcome of liver transplants by uncovering the delicate details concealed in the arterial blood pressure signal that remain indistinguishable to the naked eye.