Hi everyone. Like how “AbstractDifferentiation.jl” serves as a common interface for differentiation and enables quick testing of different backends, does anything similar exist for estimation of parameters/functions in Universal Differential equations using data? I would love to be able to define an mtk-system/ODEProblem and then be able to prototype and compare all the sciml parameter/function estimation methods i.e. SciMLSensitivity, DiffEqParamEstim, DiffEqBayes, Surrogates, ReservoirComputing, DataDrivenDiffEq, DiffEqFlux and DeepEquilibriumNetworks!
Thanks everyone!
1 Like
It’s fairly hard to come up with a nice abstract interface on this because different methods can work very differently or use very different information. But the closest to a common interface is DyadModelOptimizer, but in order for that to work it needs the ModelingToolkit.jl symbolic knowledge of the model, whereas SciMLSensitivity says “go ham on your own”, DiffEqParamEstim wants general diffeq models, etc.
2 Likes
Thank you for the answer! It makes sense that these things are hard all things considered! I’ll check out DyadModelOptimizer!