Module juice::layers::common::linear

Applies a linear transformation to the input data y = a * x + b

The variables are:

• y: output value
• a: weight (a trainable weight in a neural network)
• x: input value
• b: bias (only for Backends with the coblas::plugin::Copy trait)

Input Data

The input can either have one or two dimensions:

• If the input has one dimension the transformation will just be applied to the input data.
• If the input has two dimensions the first dimension is treated as batch size (N) and the transformation will be applied to every vector in the second dimension, using the same weights and biases.

In the context of convolutional neural networks this layer is also called a “fully-connected layer” if it is used at the end of the network.

Structs

• Linear
  Linear Layer
• LinearConfig
  Specifies configuration parameters for a Linear Layer.