This is small example of the Multi Layer Perceptron implemented in Java.

It can be used for educational purpose or experiments on this kind 
of neural networks. 
In this example the neural network will learn the XOR logical gate.

This is based on some practical exercises by Jean-Baptiste Mouret
teacher at the Institute of Intelligent Systems and Robotics in France:
ISIR, "Université Paris 6"
http://www.isir.upmc.fr/

Jean-Baptiste Mouret's personal page at the ISIR can be found at:
http://www.isir.upmc.fr/index.php?op=view_profil&id=72&old=N&lang=en

- the 1st layer, namely the input layer is composed of 2 neurons, 
- the 2nd layer, an hidden layer has 6 neurons,
- the 3rd layer, namely the output layer has 1 neuron.

The Neuron class represents a neuron of the neural network.
The Layer class represents a layer of the perceptron.
The Mlp class represents the perceptron.


To compile, use:
$ javac -g Mlp.java

Then run Mlp:
$ java Mlp


A gnuplot.dat file will be generated, it contains the plot datas for
the evolution of the quadratic error.

To plot it with gnuplot, for example:
$ gnuplot
gnuplot> plot "plot.dat" using 1:2 with lines
gnuplot> exit


You can try to adjust the number of layers, neurons, the learning rate
or the learning cycles and watch the learning accuracy.