DepthwiseConv1DLayer<T>

Layers · AiDotNet.NeuralNetworks.Layers

A depthwise 1-D convolution over [B, C, T] feature/waveform data: each input channel is convolved with its own temporal filter (no cross-channel mixing), producing C * multiplier output channels.

Inheritance: LayerBase<T> → DepthwiseConv1DLayer<T>

Implements: IShapeContract

How It Works

This is the time half of the 1-D time-channel separable convolution used by NVIDIA QuartzNet / Citrinet (Majumdar et al., 2021). A full TCSConv is this depthwise-temporal layer followed by a pointwise (1×1) Conv1DLayer that mixes channels — factorizing a dense 1-D convolution into far fewer parameters while preserving the receptive field.

Built on the tape-aware DepthwiseConv1D engine primitive, so the gradient flows through the recorded op with no hand-written backward. The kernel width is C (channels are known at construction), so weights are allocated eagerly — avoiding the lazy-init/Clone-via-SetParameters disagreement that afflicts input-inferred layers.

Constructors

Constructor Summary
DepthwiseConv1DLayer(int channels, int kernelSize, int multiplier, int stride, int? padding, IActivationFunction<T> activation, IInitializationStrategy<T> initializationStrategy) Constructs a depthwise 1-D convolution.

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)
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).