Artificial Intelligence By Example

The output layer

The output layer will now receive the result y = (input * weights+ bias) squashed by logistic sigmoid (y), as shown in the following diagram:

A dataflow graph with output layer

The output layer takes the y output of the hidden layer (on the right-hand side of the graph) and receives the weights and bias to apply to this layer in this model. Once the multiplication of weights and addition of bias are computed, the logistic sigmoid function squashes this output as well. This is a very nice way of keeping values homogeneous over the network. It's a tidy way of always sending small computable values to the next layer as implemented ...