Compensatory Learning
- I haven't seen anyone else who has done it, so I think we're rather
atypical in using compensatory rules.
- The idea is that each neuron has an amount of syntactic strength.
- If it's below the target strength and increasing, you increase by more;
below target and decreasing, decrease less; above strength and
increasing, increase less; and above strength and decreasing, decrease
less.
- This makes syntactic strength a commodity, and reduces the problems
of one or a small number of neurons taking over.
- Note that synaptic strength here can mean synapse leaving a neuron
(post-compensatory), entering (pre-compensatory) or a combination.
- We've explored both pre and post-compensatory learning, but not
the combination.
