-
-
Notifications
You must be signed in to change notification settings - Fork 232
Description
Component arrays got tags for each element. This could be used by modelling toolkit to determine individual parameters and initial condition values (in a similar way as done for e.g. a map).
While this does not directly add anything, Lux uses ComponentArray
s to represent neural network parameters. If MTK could use these, it should be relatively strightforward to work with UDEs in MTK (as one could simply merge the ComponentArray
representing neural network parameters and those representing otehr parameter values and initial conditions. Next, Optimization.jl can run directly using ComponentArray
s as well, so these could be used unmodified in the loss function.
This or something equivalently convenident might actually already by possible, but if so, I am not aware of it.