The backpropagation model can be conceptually represented as follows:
Figure 4.16: Backpropagation model
The backpropagation algorithm can easily be implemented in a staged manner. This is computationally less demanding compared to the gradient descent:
- Initialize the model: In this step, the model is randomly initialized to a point where the weights are selected with mathematical approximation and randomness. This is the first step in the feed-forward network.
- Propagate forward: In this step, all the input units, hidden units, and the output units are activated after adding the sum of the products of the neuron units ...
