Model identifiability in complex systems
by Mario Castro
Institute for Research in Technology (IIT),
ICAI Engineering School, Universidad Pontificia Comillas, Madrid, 28015 (SPAIN)
Summary:
This course is designed for students engaged in statistical physics and complexity science, offering a comprehensive exploration of critical concepts and methodologies that underpin the understanding and analysis of complex systems. The session will delve into the nuances of model identifiability, distinguishing between structural and practical identifiability and their pivotal roles in ensuring the robustness and reliability of model predictions. We will expand on traditional model fitting techniques, moving beyond least squares to incorporate sensitivity analysis, Fisher information, and Bayesian methods. We will provide participants with a toolkit to enhance model precision and inference. Furthermore, the seminar will address the challenge of "sloppy models" - models with parameters that have negligible impact on the output- and discuss strategies to recognize and mitigate their influence on the modeling process. We will use many examples from different fields to illustrate the topics and encourage the participants to program their codes to understand them deeply.
Slides and lecture recordings
