SnakeLayer<T>
Layers · AiDotNet.NeuralNetworks.Layers
The Snake activation (Liu, Hartwig and Ueda 2020): f(x) = x + (1/α) sin²(αx) with a trainable frequency α per
channel — a periodic nonlinearity BigVGAN uses to give its generator a periodic inductive bias.
Inheritance: LayerBase<T> → SnakeLayer<T>
Implements: IShapeContract
For Beginners
An activation that wiggles: it is the identity plus a learned sine-squared ripple, which helps a network produce and extend periodic signals such as the harmonics of speech.
How It Works
α starts at 1 and a small constant (1e-9) keeps 1/α finite, as BigVGAN’s Snake does. Channels are axis 1
of an input [batch, channels, time] (or axis 0 of [channels, time]).
Constructors
| Constructor | Summary |
|---|---|
SnakeLayer(int channels) |
Creates the layer. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<string, string> GetParameterRoles() |
Returns parameter roles for per-role optimizer configuration (e.g., weight decay exemption for biases). |
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
Derived from this type’s [TensorLayout] declarations. |
IReadOnlyList<Tensor<T>> GetTrainableParameters() |
Returns all trainable parameter tensors marked with [TrainableParameter]. |
void ResetState() |
|
void SetTrainableParameters(IReadOnlyList<Tensor<T>> parameters) |
Replaces trainable parameter tensors (e.g., with ParameterBuffer views). |
void ZeroGrad() |
Clears all gradient fields discovered by convention ({paramName}Gradient). |