MultiFidelityPINNOptions

Options & Configuration · AiDotNet.PhysicsInformed.Options

Configuration options for the MultiFidelityPINN.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → PhysicsInformedOptions → MultiFidelityPINNOptions

For Beginners

A multi-fidelity PINN learns from two sources at once: a lot of cheap, rough data and a little expensive, accurate data. The weights here decide how much it trusts each, and how hard it tries to keep the two consistent.

Constructors

Constructor Summary
MultiFidelityPINNOptions() Initializes a new instance carrying the model’s published defaults.
MultiFidelityPINNOptions(MultiFidelityPINNOptions other) Initializes a new instance by copying another instance.

Properties

Property Summary
bool FreezeLowFidelityAfterPretraining { get; set; } Gets or sets whether the low-fidelity network is frozen once pretraining ends.
double BoundaryWeight { get; set; } Gets or sets the weight on the boundary condition loss term.
double CorrelationWeight { get; set; } Gets or sets the weight on the term tying the two fidelities together.
double HighFidelityWeight { get; set; } Gets or sets the weight on the expensive, scarce high-fidelity data.
double LowFidelityWeight { get; set; } Gets or sets the weight on the cheap, plentiful low-fidelity data.
double PdeWeight { get; set; } Gets or sets the weight on the PDE residual loss term.
int NumCollocationPoints { get; set; } Gets or sets how many interior points the PDE residual is evaluated at.

Methods

Method Summary
void Validate() Throws if a value this model requires has been left unset or is not positive.