Learning

There's not much learning in this simulation, however it is needed for variable binding.
Learning is strictly Hebbian, which means that synaptic weights are changed based solely on the properties of the pre and post-synaptic neurons.
The typical Hebbian rule increases the weight when the neurons fire simultaneously and decreases the weight when only one fires.
Most of our work has used a rule that reduces weights when the presynaptic neuron fires and the postsynaptic neuron does not.
We use a compensatory Hebbian learning rule.
This sets a goal weight for total synaptic strength leaving a neuron.
If the total strength is less then the goal then more weight is added during increase, if the total strength is less, then less strength is added. Reduction follows the reverse pattern