Hands-On SCiML - SINDy

Bring your laptop to the hands-on ML session on SINDy, run by Alasdair Roy and hosted by the SciML community! Alasdair will give a 45-minute introduction to SINDy. After a break for coffee and cake, we'll tackle the coding challenges that Alasdair has prepared for us.

(Note: This is an in-person event. Only the first 45 minutes will be streamed/recorded.)

Bio:

I am a PhD student in the CDT for fluid dynamics working on reduced-order modelling methods using the sparse identification of nonlinear dynamics (SINDy). In my research, I use SINDy to identify dynamical systems from magnetoconvection simulation data by seeing what characteristics of the data we are trying to reproduce.

The hands-on-ML session:

In the presentation, I will talk about the SINDy method and the accompanying Python package (PySINDy). SINDy is a sparse regression package, which allows symbolic identification of dynamical system from time series measurements. This makes SINDy different to black box methods, as finding underlying governing equations of motions allows for physical interpretation of the system, potentially allowing for greater generalisability of the resulting model. We will look at how the original method works along with several relevant extensions as well as some of their weaknesses. Some extensions will include: WeakSINDy for identification in noisy data and identification of PDEs, SINDy-PI for implicit model identification, choice of optimisers and including known constraints. In the coding session, we will cover some of the methods discussed above by looking at a case of ODE and PDE discovery using PySINDy.