Skip to content
ML Academy

Introduced on day 31

Dense (fully connected) layer

A layer in which every unit sees every input, computed as H = activation(XW + b) with W of shape (inputs, units) and one bias per unit.

In detail

For a batch of B rows, X has shape (B, n_in), W has shape (n_in, n_out) and b has shape (n_out,); the output has shape (B, n_out). The layer has n_in * n_out + n_out parameters, whatever the batch size. Convolutional and recurrent layers share weights differently and have different counts.