A Range of Hebbian Learning: Compensatory Learning
- There are a whole range of options within the
Hebbian learning rule.
- We have added a compensatory modifier.
- Each neuron has a desired total synaptic weight
(the sum of all synapses leaving a neuron).
- If the total is below this goal weight, more weight is
added during learning, and less taken away when forgetting.
- Conversely, if the weight is above the goal, less weight is
added during learning, and more taken away during when
forgetting.
- This prevents particular neurons from having too much effect,
and helps all neurons to participate.
- Compensatory learning is needed for neural recruitment with
spontaneous activation.
- It also allows sparser patterns to form CAs.
