ga-nn-ag - Summary
This group is not part of the GNU Project.
The project is a tool that searches for an optimum neural network employing a genetic algorithm as a refiner.
The neural network defines connections in a neuron to neuron scheme, instead of layer to layer (which is the common way in feedforward networks), making easier to add / remove neurons and connections.
Almost every single parameter of the neurons could be manipulated by the genetic algorithm, since almost every parameter is defined in a per neuron basis.
Two parameters controls the evolving networks: size and error.
Size is the number of neurons that are present in a network. This parameter could be deactivated.
Error is the square of maximum error at the output stage (it could easily be changed to the mean square of the cumulative error).
The system is designed in a client-server model, so multiple neural networks (the clients) could be trainned and tested in parallel.
<meta keywords="neural network feedforward">
Registration Date: Sun 05 Sep 2004 09:25:42 AM UTC
License: GNU General Public License v2 or later
Development Status: 4 - Beta
- Browse open items
Submit a new item
- Browse open items
Submit a new item
Powered by Savane 3.13-caa5.
Corresponding source code