 |
 |
   |
|
|
|
To solve problems which are not linearly separable, the idea is to use combinations of neurons. The outputs of the ones are used as inputs of others. For example, in the case of exclusive OR the following combination can be assumed to perform an exclusive OR:
Figure 29: Perceptron and exclusive OR
The large majority of networks are organized in successive layers, the outputs of a layer are used as input for the following layer. The difficulty consists in finding the best configuration adapted to the current problem. The following figure shows a layer of M neurons with N inputs:
Figure 30: Example of layer of neurons.
The computation of the outputs can be expressed in a matrix
form as follows:
The functions of activations are applied, if necessary, to the components of Y.
