Fitting a dynamic system with an exogenous input (nonhomogenous neural ode) via DiffEqFlux

[Volker]

Thank you very much

Sure you can use it for the documentation.

@ChrisRackauckas

so you were broadcasting a row vector against a column vector in the loss function to generate a matrix.

Oh no, I haven´t seen this. So simple… This was probably the main reason, why it didn´t converge

@ChrisRackauckas

If you don’t have very much data but you have a model then this could be a way to get a better predicting method since the neural network will likely overfit and attempt to over-generalize. There’s also a ton of other reasons, like how discretized physics-informed neural networks are a subset of UDEs and how high dimensional PDE solvers can be written as a UDE. See https://arxiv.org/abs/2001.04385 for all of the details.

Thanks, I will read it carefully.