Computations in Science Seminars
May 2018
23
Wed 12:15
Niall M. Mangan, Northwestern University
e-mail:
Host: Arvind Murugan ()
Organizer: Yuval Yifat ()
Identification of Hybrid Dynamical Systems via Clustering and Sparse Regression

Inferring the structure and dynamical interactions of complex systems is critical to understanding and controlling their behavior. Hybrid systems are challenging to identify because the parameters and equation structure may change across multiple dynamical regimes. Key examples include varying transmission rates in epidemiological problems and legged locomotion. Many current methods focus on inferring a model for the system and detecting switching points in a time-centric framework. One can reframe the problem by clustering in data-driven coordinates, such that similar dynamical behavior is close together, and then use the sparse identification of nonlinear dynamics (SINDy) method to identify different dynamical regimes. I will discuss model selection using SINDy and information criteria. I will then demonstrate the success of the method hybrid-SINDy on a spring-mass model and a simple infectious disease model with time-dependent transmission rates. I will also investigate robustness to noise and cluster-size.

May 2018
30
Wed 12:15
Pedro J Saenz, MIT
e-mail:
Host: William Irvine ()
Organizer: Delphine Coursault ()
Jun 2018
6
Wed 12:15
OPEN