Expand description
Dense layer and element-wise activations for deep kernel feature extractors.
A DenseLayer is an affine transformation y = Wx + b followed by
an element-wise Activation. The layer owns row-major weights
(weights[i][j] is the contribution of input j to output i) and a
parallel bias vector. The forward pass is plain f64 arithmetic — we
deliberately stay out of ndarray here to keep the module lightweight
and easy to reason about for gradient derivations.
Structs§
- Dense
Layer - An affine-plus-activation layer
y = activation(Wx + b).
Enums§
- Activation
- Element-wise activation applied at the output of a
DenseLayer.