i started looking into neural networks, and so far ive learnt feed foward networks with back propogation supervised learning
to test it out i did various things, but this little test i do like alot:
the function being modled is 0.25(sin(pi.x)cos(pi.x)+1.5)(2-x)
with x from 0->2, the neural network is using sigmoid threshold functions, so values of y have been made to be in the range 0-1 for the network to work:
every frame, it runs through 100 times with each black dot you see on the screen training the network, and updating a graph using the network to get the value of y for corresponding x, you can see that pretty quickly it converges to the functions exact graph.
the network is pretty small, 1 input, 6 hidden neurones, 1 output neurone, and im using my own class that i built for a generic feed foward neural network using back propogation supervised learning so you can have a tiered network, lets say:
12 inputs, 5 hidden units, 6 hidden units(another layer), 4 hidden units (another layer), 5 output units.
http://denvish.net/ulf/1169652114_neuronegraph.php (requires flash player 9)
my next point of research, is into unsupervised learning, with the ultimate goal of making a computer learn how to play some sort of simple game like tic tac toe or sometihng
The Beginning Of All Things To End.
The End Of All Things To Come.
I love the subject! I wrote some NN programs a few years ago (in C).
I built a few classes in what I called "Entity World". I wanted several NN entities, with inheritable characteristics (using Genetic Algorithms).
My idea was to set up a simple environment and have their survival based on how well they adapt.
So, instead of a game, how about seeking light? You can start with light at the top half. Then when they learn to rise, start moving the light, etc. Just like primitive life (he says, with a Frankenstein grin).
(BTW the animation is cool.)
What are you using to develop in Flash Player 9?