public override backwardTraining(context as LayerPass<T>, upstream as NDArray<T>) as LayerGradients<T>
Differentiates the saved affine result without repeating its matrix product.
| Parameter | Description |
|---|---|
context | Forward context from this layer. |
upstream | Output derivative, including the activation. |
Input and ordered parameter derivatives.
gradients = layer.backwardTraining(context, upstream)