The Interactive Neural Network (The INTER-NET!)
Introducing the interactive neural network!
You can play around with it here:
Shape your network with the sliders below, adding biases, different activation functions, inputs outputs layers and hidden units per layer.
This is really just a toy to explore how neural networks work at a low level, where you can see each connection and which neurons are changing the most as you change the input. So far I've had some cool "instructive moments" where I learned something new about neural nets that I wouldn't have seen otherwise!
Then drag the neurons to any value you'd like! When you drag a neuron, you're introducing a small change to its value, which is back propagated to all weights and parent neurons of said neuron.
If you'd like the computer to train for you, drag inputs and output neurons to your desired values, and double tap a letter on your keyboard like "a". Now whenever you hit "a" it will bring those neurons to the values you dragged them to before. Save a few more examples under different letters "b" and "c" (or whatever you like, but not "t"). Then hold "t" to train your network. It will pick random training examples and pull the outputs to the target values you specified.
I'm sure there are a few bugs, if you find any let me know in the comments and I'll fix em up!
If you want to see the code it's on github here: https://github.com/JonComo/INTER-NET