SourceForge.net Logo About AINN
 Support This Project
[Home]

 

Background

Willing to develop a simulation game, I needed to access some well-done oriented object C++ classes of different neural network types (mainly the back propagation and the bidirectional associative memory).

After some long search in the Web, I managed to find exactly what I needed: a simple easy to use (do not need a 300 pages book before usage) and so-called object oriented...

Well! I experienced some frustrations!

First, I have noticed that most of the people who were developing the neural networks have few concern for the object oriented coding. Most of the time the neural network object is concentrated in one big huge unreadable function that nearly does it all. The only exception was with the "Object Oriented Neural Networks in C++" from J. Rogers - but again that was to my sense going a little too far on the other side - although excellent for neophytes in Object Oriented C++ programming.

Second, few of the coding include a function to avoid local minima and when it does, the functionality is merely to justify a mathematical method (annealing or genetic algorithm) but not what I would consider as really practical.

Third, most of the library are for free during a short period of time. Then you will have to purchase it. I find it a shame that knowledge should/would be sold in the market - it is for me a universal gift and should be available to all - (do not mix up with application that are using knowledge)

There must be a correct object oriented neural network library somewhere!!!

I have to find the time to find it and when it is found, it must be free of charge... (can still hope to dream)

This is nearly impossible to find, this is why I decided to open the sources of my own object oriented neural network library. At least I know how it is engineered how I like it and does what I expect it to do!

Purpose

I want to make available to all  a neural network library object oriented. This library will contain different neural network types and each of them would easily combine to build up hybrid neural networks.

Each neural network will published with the documentation and some example test coding. Also, on time, I intend to write some papers summing up the result and what I have found or thought.

It is also the purpose to clear the fog that encapsulate the neural network theory and make it available to everyone.

Also the site will offer you a series of interesting page links to other sites with also useful information.