Creating Neural Networks

 

Creating a Backpropagation neural network

1) Select New->Backpropagation Neural Network from the file menu .

2) The New Backpropagation Neural Network dialog will load.

3) Input settings and hit "OK". What do all these settings mean? 

Input Layer Size: The number of neurons in the input layer. Should be the largest of the three layers.

Hidden Layer Size: The number of neurons in the hidden layer. Should be smaller than the input layer and larger than the output layer.

Output Layer Size: The number of neurons in the output layer. Should be the smallest of the three layers.

Input Size: The number of inputs going into the input layer. 

Punish Rate: The rate at which link weights (links between neurons) are punished if they are causing an incorrect output for a particular instance of training data.

Reward Rate: The rate at which link weights (links between neurons) are rewarded if they are causing a correct output for a particular instance of training data.

Error Threshold: The level of error that a neuron must be under to stop training. Do not set this to or below zero. Neurons will have a certain level of error even when the network is producing correct results.

Minimum/Maximum Initial Link Weight: In a Backpropagation neural network, the link weights (links between neurons) are initially assigned a random value. These two fields set the range that these weights will initially fall into. As the network trains it is possible, sometimes even likely, that the link weights will grow outside of this initial range. 








< Neural Networks

Loading/Saving >