PANNsPoolingLayer<T>
Layers · AiDotNet.NeuralNetworks.Layers
Implements CNN14’s channel-preserving frequency/time aggregation for NCHW features.
Inheritance: LayerBase<T> → PANNsPoolingLayer<T>
Implements: IShapeContract
How It Works
The released PANNs CNN14 first averages the mel-frequency axis, then adds the maximum and mean across time. Engine reductions are used so gradients remain on the training tape. It is an architectural detail of the PANNs factory rather than a general channels-last global-pooling variant, but it is public like every other layer so a caller can compose or subclass it directly.
Constructors
| Constructor | Summary |
|---|---|
PANNsPoolingLayer() |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
|
void ResetState() |