NonlinearROMs.jl

Neural-network-based hyper-reduction and nonlinear reduced-order modelling components for GridapROMs.jl.

This package was extracted from GridapROMs.RBSteady/GridapROMs.RBTransient (the NonlinearModels.jl, NNHyperReduction, NNOperatorReduction, and HighDimNN* transient counterparts) into its own repository, and plugs back into GridapROMs via multiple dispatch — no changes to GridapROMs itself are required.

Installation

NonlinearROMs is not a registered package, so it must be added as a dev dependency (or otherwise made resolvable) before instantiating this environment:

using Pkg
Pkg.develop(path="../NonlinearROMs.jl")
Pkg.instantiate()

Usage

using GridapROMs
using NonlinearROMs

res_reduction = NNHyperReduction(tol; nparams, sketch, compression)
jac_reduction = NNOperatorReduction(tol; nparams)
rbsolver = RBSolver(fesolver, state_reduction, res_reduction, jac_reduction)

NNHyperReduction/NNOperatorReduction (and their transient counterparts HighDimNNHyperReduction/HighDimNNOperatorReduction) can be passed anywhere a HyperReduction is expected, exactly like MDEIMHyperReduction or RBFHyperReduction.

Contents

• NeuralNetworks.jl — MultiLayerPerceptron, GenericNeuralNetwork, AutoEncoder, VariationalAutoEncoder, AutoDecoder, train_neural_coefficient, NNStrategy.
• SteadyReductions.jl / SteadyHyperReductions.jl / SteadyInterpolations.jl / SteadyReducedOperators.jl — steady NNOperatorReduction/NNHyperReduction.
• TransientReductions.jl / TransientHyperReductions.jl / TransientInterpolations.jl / TransientReducedOperators.jl — transient HighDimNNOperatorReduction/HighDimNNHyperReduction.

Testing

using Pkg
Pkg.test("NonlinearROMs")