One of the most commonly used networks is the multilayer feed forward
network. It falls under the category, "Networks for Classification
and Prediction" and has widespread interesting applications and functions.
We have isolated this specific network category as the neural network
we have built ourselves (conceivable word checker), makes use of the
principles pertaining to this type of network. In this section of
the project, we aim to discuss multilayer feed-forward networks, their
history, basic architecture, training and such a network as
well as their uses and applications.
Feed-forward networks are advantageous as they are the fastest models to execute, and as mentioned before are universal function approximators. One major disadvantage of this network type, is that no fast and reliable algorithm has yet been designed and therefore can be extremely slow to train . Thus, in conclusion multilayer feed-forward networks should be chosen if rapid execution rates are required, but slow learning rates are not a problem. (Masters,1993, pg 116)