Struct SpectralNormalization 

pub struct SpectralNormalization {
    pub target_norm: f64,
    pub lambda: f64,
    pub power_iterations: usize,
}

Spectral Normalization regularizer.

Normalizes weight matrices by their spectral norm (largest singular value). Useful for stabilizing GAN training and improving generalization.

References

• Miyato et al. (2018): “Spectral Normalization for Generative Adversarial Networks”

Fields

target_norm: f64

Target spectral norm (usually 1.0)

lambda: f64

Strength of the regularization

power_iterations: usize

Number of power iterations for spectral norm estimation

Implementations

impl SpectralNormalization

pub fn new(lambda: f64, target_norm: f64, power_iterations: usize) -> Self

Create a new spectral normalization regularizer.

Trait Implementations

impl Clone for SpectralNormalization

fn clone(&self) -> SpectralNormalization

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SpectralNormalization

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for SpectralNormalization

fn default() -> Self

Returns the “default value” for a type.

impl Regularizer for SpectralNormalization

fn compute_penalty( &self, parameters: &HashMap<String, Array<f64, Ix2>>, ) -> TrainResult<f64>

Compute the regularization penalty for given parameters.

fn compute_gradient( &self, parameters: &HashMap<String, Array<f64, Ix2>>, ) -> TrainResult<HashMap<String, Array<f64, Ix2>>>

Compute the gradient of the regularization penalty.