GatedFusionLayer<T>
Layers · AiDotNet.NeuralNetworks.Layers
Gated fusion of two equally-wide feature streams, as used by ABINet’s fusion stage.
Inheritance: LayerBase<T> → GatedFusionLayer<T>
Implements: IShapeContract
For Beginners
Two components each produce a guess. Rather than averaging them or always trusting one, this learns a per-value dial: where the dial is near 1 it takes the first component’s answer, near 0 the second’s, and in between it blends them.
How It Works
Implements ABINet’s gated combination (Fang et al., CVPR 2021, arXiv:2103.06495 §3.4):
The gate is learned from BOTH streams jointly, then used to interpolate between them per position and per channel — so the model can lean on the visual reading where the image is clear and on the language model’s correction where it is not.
Input is the two streams concatenated along the last axis, [..., 2 * width], which
keeps the layer inside the single-input ILayer contract. Output is
[..., width].
Constructors
| Constructor | Summary |
|---|---|
GatedFusionLayer(int width) |
Creates a gated fusion layer. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<ILayer<T>> GetSubLayers() |
|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
|
void ResetState() |
|
void UpdateParameters(T learningRate) |