Neural Network Software

PracticalStudies.com Neural Network Software

A class library for constructing artificial neural networks.

See:
          Description

Packages
com.practicalstudies.neural Neural network class library.
com.practicalstudies.neural.test Test classes for the neural network class library.

 

A class library for constructing artificial neural networks.

Networks are constructed as sets of Nodes connected by WeightedConnections. In practice, the Nodes will be represented by Neurons, suitably subclassed, using appropriate Adder and ActivationFunction objects to calculate Neuron output, and an appropriate WeightAdjustmentFunction for network learning.

For backpropagation training, use BackpropagationNode, with LinearAdder and SigmoidalActivationFunction to calculate Neuron output, BackpropagationErrorFunction to determine errors in Neuron output, and BackpropagationWeightAdjustment to train the network. On reflection, this breakdown seems overly complex, and I am looking to rationalize it.

See BackpropagationTest as an example of how to use the indicated classes.


PracticalStudies.com