Human population dynamics as a Bayesian inverse transport problem

Abstract

Many open problems across physical, biological, and engineered systems involve non-equilibrium transport processes where the governing conservation laws are known, but the underlying constitutive relations remain latent and time-varying. Conventional data-driven approaches like deep neural networks capture statistical patterns but routinely violate fundamental mass conservation. Here, we introduce a unified Bayesian inverse transport framework that resolves this by embedding Bayesian Neural Networks (BNNs) directly within exact partial differential equations in age-time space. By evaluating this framework on complex, real-world human cohort advection across China, Japan, and South Korea, we demonstrate that this physical constraint enables consistent uncertainty propagation and missing-data reconstruction from sparse observations. Beyond demography, this framework provides a generalizable foundation for observing and forecasting non-equilibrium boundary dynamics across various fields.