Algodoo

I've fallen into my machine learning phase, this time, having something to show! In this scene are neurons that don't rely on global variables, making them conveniently copyable! Every neuron has:

Dendrite - circle that recieves signals and computes an output
Axon - laser that transmits the output of the circle out to another neuron

Every neuron also includes, in it's script menu:

_acc - internal accumulator value
_bias - bias, added to _racc before being sent out the axon
_racc - "real" accumulator value, set to _acc if _acc is greater than _thr, otherwise set to 0
_thr - threshhold, a static number _acc must be greater than for _acc to be sent out instead of 0

The neural circuit in this scene is a sort of delay line memory, able to hold either a static high or low value, or, if you can pull it off, it can hold an oscillation between the two. Dark red means inactive, neutral, green means positively active, blue means negatively active.
Use keys WS to excite and tame the circuit. Press D to forget a portion.
The graph displays what value passes through a specific neuron in the delay line, and repetition can be seen, as well as how the information evolves over time. Notice how an oscillation of positive-negative usually collapses into a static positive. I'm not sure why the circuit has a tendency to grow positive portions and shrink negative ones, but it does so anyway.
Added new types: Atan, Boundless, Binary, Trinary Neurons