Structural identification of dynamical systems -- overview and reflection
Dag Wedelin, Chalmers University
Abstract:
We consider how to algorithmically identify both structure (network of interactions, form of ordinary differential equations), and parameters of dynamical systems from time series data, in particular for biomolecular systems modeled with chemical rate reactions. I will outline current research and challenges in this field, illustrate how the problem can be approached by algorithms operating locally in the network structure, and relate to other similar problems such as function identification and structural learning of probabilistic graphical models. Finally, I will discuss some cross-disciplinary issues with respect to objectives, problems and methods, and the general prospects of algorithmic model selection.