Nikolaos Sidiropoulos, Louis T. Rader Professor of electrical and computer engineering at the University of Virginia, and his collaborators have developed models and methods to predict the evolution of epidemic trends for many regions simultaneously. This helps public health officials and hospital administrators manage scarce resources such as respirators and intensive care unit beds needed for COVID-19 response.
Their method predicts future hospitalizations by finding meaningful relationships among three variables: state- or county-level geographic location; different medical indicators such as new infections; and time. The key idea in their method is to use latent instead of location/attribute-level epidemiological dynamics to capture common epidemic profile sub-types and improve collaborative learning and prediction. Their experiments using both county- and state-level COVID-19 data show that the proposed model can identify interesting latent patterns of the epidemic and predict the number of hospitalizations 10-15 days ahead with accuracy up to 25% better than the best available baselines. This is joint work with IQVIA’s Nikos Kargas, Cheng Qian (formerly a postdoctoral researcher in Sidiropoulos' group at UVA), Cao Xiao and Lucas Glass, and Jimeng Sun, professor of computer science at the University of Illinois Urbana-Champaign. They will present their paper, STELAR: Spatio-temporal Tensor Factorization with Latent Epidemiological Regularization, at the Association for the Advancement of Artificial Intelligence annual conference February 2-9.