torch.js has not been released yet.

torch.nn.utils.spectral_norm

function spectral_norm<T extends Module>(module: T, options?: SpectralNormOptions): T

function spectral_norm<T extends Module>(module: T, name: string, n_power_iterations: number, eps: number, dim: number, options?: SpectralNormOptions): T

Apply spectral normalization to a parameter in the given module.

Spectral normalization constrains the Lipschitz constant of the layer by normalizing the weight by its spectral norm (largest singular value). This is commonly used in GANs to stabilize training.

The spectral norm is estimated using power iteration.

Parameters

moduleT: Module containing the parameter
optionsSpectralNormOptionsoptional: Optional settings for spectral normalization

Returns

T– The module with spectral normalization applied

Examples

const conv = new torch.nn.Conv2d(3, 64, 3);
torch.nn.utils.spectral_norm(conv, { name: 'weight' });

// The weight is now normalized by its spectral norm

See Also

PyTorch torch.nn.utils.spectral_norm()

SpectralNormOptions