Conceptual

Practical Identifiability Analysis of Dynamic Models via the Fisher Information Matrix

How to decide whether the parameters of a dynamic (ODE/PDE) model can be reliably estimated from noisy data. A rigorous definition of practical identifiability is shown to be equivalent to invertibility of the Fisher Information Matrix, connected to coordinate identifiability from the profile likelihood for a cheaper metric; for singular-FIM cases, eigenvalue decomposition finds non-identifiable directions that are handled with regularization, and an optimal experimental design algorithm chooses data collection so every parameter becomes identifiable.