Summary
It would help with Physics-enchanced Neural ODEs to leverage the existing NeuralNetwork 3.0 library.
If this was natively supported by the IR's (similar to what we are doing with PDE stencils) this could map to Cuda etc through the Solve IR.
Context
https://github.com/AMIT-HSBI/NeuralNetwork
https://elib.dlr.de/200100/1/TKAMP_2023_ICINCO.pdf
https://openmodelica.org/images/M_images/OpenModelicaWorkshop_2025/2025-02-03_phannebohm_NeuralNetwork.pdf
Expected Behavior
Compiling a modelica model with a neural net should generate a model ready for SciML workflows.
Summary
It would help with Physics-enchanced Neural ODEs to leverage the existing NeuralNetwork 3.0 library.
If this was natively supported by the IR's (similar to what we are doing with PDE stencils) this could map to Cuda etc through the Solve IR.
Context
https://github.com/AMIT-HSBI/NeuralNetwork
https://elib.dlr.de/200100/1/TKAMP_2023_ICINCO.pdf
https://openmodelica.org/images/M_images/OpenModelicaWorkshop_2025/2025-02-03_phannebohm_NeuralNetwork.pdf
Expected Behavior
Compiling a modelica model with a neural net should generate a model ready for SciML workflows.