ModelingToolkitNeuralNets

[AlCap23]

ModelingToolkitNeuralNetworks

This would be nice to include in the near future ⌛

[AlCap23]

@ChrisRackauckas is the API of MTKNN able to produce gradients and hessians w.r.t. to the parameters right now?

[ChrisRackauckas]

It's just a model builder. If the code it builds is differentiable then it is, and it just uses Lux and SciML solvers the standard way. There's examples of the generated models being fit in the docs.

[AlCap23]

Ah, sorry I wasn't clear. Symbolic gradients and hessian 😅. I saw the wrapper , but was still unsure about the symbolic derivative part.

We've been exper with this using FastDiff right now, but it seems a customised chain rules / diff rule for the full network works better.

[ChrisRackauckas]

No matrix calculus in symbolics right now