FEED FORWARD NEURAL NETWORKS - A JAVA IMPLEMENTATION v2.0 by Aydin Gurel

What is this? This is a java implementation of some types of feed forward neural networks. You can simulate multilayer perceptron nets, generalized feed forward nets, modular feed forward nets using this package. You can use various back propagation methods. The source code is available and free.

What can you do with it? You can: - Build multilayer perceptron, generalized feed forward, modular feed forward nets; - Train your net using batch, mini batch and incremental training; - Create pattern sets out of semicolon separated files, train the net using these patterns, - Spare some of the patterns for cross validation and testing, calculate error terms out of these patterns; - Load and save network configuration and weights; - Use logistic, tanh and linear activation functions; - Use momentum; - Determine the flatness of the activation functions; - Use different momentum rate, flatness, learning rates for each neuron.

More information Please see FAQ for more detailed information.

Examples There are three examples and each of them demonstrates different features of the package. - In example 1; we create a generalized feed forward net out of a configuration file and we train it using a set of patterns. We spare some of the patterns for cross validation and test. - In example 2; we create a multilayer perceptron and train the net using a set of patterns. - In example 3; we create a simple multilayer perceptron and train it incrementally without using pattern sets.

Download all in zip format (source codes and documents)

The code is free, but please contact me if you wish to use the code entirely or partially in any kind of project so that I can reference it and please don't delete the top lines so that other people can reach this information. Also, please inform me if you encounter a bug.

Previous versions v1.1   v1.0

FAQ | Example 1 | Example 2 | Example 3 | Download | | v1.1 | v1.0

.