Fig. 1: The proposed HOIST model.
From: Evidence-driven spatiotemporal COVID-19 hospitalization prediction with Ising dynamics
We use static data to calculate the latent distances between locations. The latent distances are then used to calculate the adaptative connectivity of the location graph. The dynamic data are used to estimate the external fields (EFs) and then generate predictions using the long short-term memory network (LSTM). We use the Ising dynamics to regularize the spatiotemporal prediction results using the estimated external fields (EFs) and the adaptative connectivity.
