Struct NeuralOde 

pub struct NeuralOde<F: OdeFunc> { /* private fields */ }

A Neural ODE layer that wraps an OdeFunc with fixed integration limits.

The forward pass integrates y(t0) = y0 to y(t1) and the adjoint method computes dL/dy0 and dL/dθ via the adjoint sensitivity method without storing all intermediate activations.

Implementations

impl<F: OdeFunc> NeuralOde<F>

pub fn new(func: F, t0: f64, t1: f64) -> Self

Create a new NeuralOde with default solver configuration.

pub fn with_config(func: F, t0: f64, t1: f64, config: OdeSolverConfig) -> Self

Create a new NeuralOde with a custom solver configuration.

pub fn forward( &self, y0: &[f64], params: &[f64], ) -> Result<OdeSolution, OdeError>

Forward pass: integrate y0 from t0 to t1.

Uses the adaptive DOPRI5 solver internally.

pub fn adjoint( &self, y0: &[f64], params: &[f64], grad_output: &[f64], ) -> Result<AdjointResult, OdeError>

Full adjoint sensitivity method.

Integrates forward to obtain y(t1), then runs the augmented backward ODE to compute dL/dy0 and dL/dθ in a single backward pass.

Arguments

• y0 – initial state.
• params – ODE parameters.
• grad_output – upstream gradient dL/dy(t1).