Multilayer neural network (perceptron) is a neural network consisting of an input, output, and one or several hidden layers of neurons.
To build and train a multilayer perceptron, it is necessary to select its parameters according to the following algorithm:
- Determine the meaning of the input vector X components. The input vector must contain the formalized condition of the problem,i.e all information necessary to get an answer.
- Select the output vector Y so that its components contain the complete answer for the task.
- Select the activation functions of neurons. It is important to take into account the specifics of the problem, since a good choice of activation functions will increase the speed of learning.
- Select the number of layers and number of neurons in the layers.
- Set the range of input, outputs, weigths and thresholds based on the selected activation functions.
- Randomly assign initial values to weights and thresholds. The initial values should not be so large that the neurons will be saturated saturated (in the horizontal part of the activation function), otherwise the learning will be very slow. The initial values should not be too small, otherwise learning will also slow down.
- Execute training, that means to select the network parameters so that the problem is solved in the best way. After training, the network will be able to solve problems of the type for which it was trained.
- Feed into trained neural network the conditions of the problem in the form of a vector X and calculate the output vector Y, which will give a formalized solution of the problem.