deep learning - sandbox

to understand how a neural network is working I'm building a neural network sandbox.

The sandbox provides a tiny network with 4(+bias) Input nodes, 5 hidden nodes and 4 Output nodes.

You can draw the connections (when editing click hold a node and drag to dest node ), delete(double click) them and set their weight (when training click hold and drag hor.) and you can set backpropagading connections.

Each neuron(node) has it's own function for value, trigger, backprop.

You can train the network (to perfection) and play with it after training to test it.

With each interation you can check the resulting values and node states to see what happens while leaning.

Here you can give it a try (current [fix4all] functions: val:passthrough, trigger:binary, backp:delta*0.1)

There will also sigmoid, tanh and other functions be available soon (and you will able to change them for each/all nodes).

link: http://www.gameus.de/nn

mercuryus

When you open the game the controls at the bottom are hidden.

I could not tell what I was supposed to be doing until I came back here and seen that the controls are there but hidden.

I like the chalk board effect and I am a big fan of educational games and using sandbox features to interact.

my fault - i changed to a better scale method.

Realod (clear cache) to get a better workplace

ps: there might by still errors (e.g. with the backpropagandation)

my fault - i changed to a better scale method.

Realod (clear cache) to get a better workplace

ps: there might by still errors (e.g. with the backpropagandation)

OK now I can see the controls. Still not sure what I am supposed to be doing but looks interesting. Maybe add an instruction screen to explain the neural network concept if this is used by people not in your class?

Thank you for your suggestions!

I will add a help screen to introduce the framework.

Understanding a neural network is not that difficult.

There are plenty of tutorials on youtube for any interested.