How to define multi-layer perceptrons (MLP)

MLP is a branch of ANNs widely used in pattern recognition because of its ability of identify patterns within noisy or unexpected environments. MLP can be used to implement supervised and unsupervised learning (both of them were discussed Chapter 9, Object Recognition). In addition to that, MLP can also be used to implement another kind of learning such as reinforcement learning inspired by behavioral psychology, where the network learning is adjusted using reward/punishment actions.

Defining an ANN-MLP consist of deciding the structure of the layers that will compose our net, and how many nodes will be in each of them. Firstly, we need to decide what the goal of our network is. For instance, we could implement an object recognizer, in which case, the number of nodes belonging to the output layer will be the same as the number of different objects we want to identify. Simulating the example from Chapter 9, Object Recognition, in the case of recognizing...