Hebbian Learning

A while ago we showed (and I don't think it's particularly surprising) that neurons are Turing complete. That is, given enough neurons (the simple IF kind or more complex ones), you can program anything.
So what's really interesting about neurons is that they learn.
They learn in lots of ways, but the simple and widely used way is permanent change of synaptic weights. In biology this is long-term potentiation (and depression).
The way we learn is a Hebbian rule:
If neuron A tends to cause neuron B to fire, the synaptic strength increases.
This of course is underspecified.
There is not wide spread agreement on the learning rule that works.
Currently spike timed dependent plasticity (STDP) is in favour.
If the pre-synaptic spike happens before the post-synapatic spike, increase the weight, if after decrease.
We started working with an older rule called Oja's rule.
It forces the synapses to the likelihood that the post-synaptic rule fires when the pre-synaptic rule does.
We called it the correlatory rule, but it's Oja's.