Overlapping CAs
- In our brains, neurons must participate in multiple CAs
- It allows more CAs (more CAs than neurons?)
- Categories overlap (my neural correlates are shaky)
- Should help with using CAs together (Associative Memory,
Hierarchies, Fast Learning)
- This means that the synaptic weights are going to have to
allow overlapping stable states (related to Hopfield nets)
- This means that synaptic weights now have less to
do with the copresentation data
- It also makes the CA simulation difficult and interesting.
- Without overlapping CAs, simulated epilepsy is easy to avoid,
because there are few connections between CAs.
- All that is needed is to make intra-CA connections very strong.
- Even a clipped simple Hebbian learning rule will work.
- With the overlapping simulation, a balance has to be struck.
