The Neural.NET Library is a .NET library for feed-forward neural networks. It includes facilities for managing and massaging the input data.
06/15/05 - Neural.NET is a neural network library for .NET. Unlike the original version of Neural.NET, this feature is a port of the Jo Dene Neural Network library. It was also enhanced to rely on the .NET event model to handle periodic events.
It implements the following features from JDNN:
I used this iplementation in a project for my robotics class, for face detection in images. To detect faces the code was extended to support convolution networks, which alternate convolution and sub-sampling layers. It was fairly easy to extend the existing source code.
To implement specific features, such as adjusting deltas or adjusting training rates while training, there is an event based mechanism to add functionality to the engine. Siimply listen for an "EpochEvent" and your handler will be notified when a training epoch completes.
This library is licensed under the LGPL, and subject to the terms and conditions of the LGPL 2.1. This means that you can dynamically link against these libraries without having to place your code under the GPL or LGPL (although you are encouraged to do so). You may freely redistribute these libraries subject to the LGPL but you must provide the source code to those libraries. You may not include the source code in a larger work (i.e. copying the code into your project) without also placing your code under the GPL or LGPL.