Long Term Dynamics
- CAs, and thus categories, are learned
- The CA is formed by changing weights of connections
- This can be done via Long-Term Potentation and Long-Term Depression
- This is done by a Hebbian learning rule
- The concept is learned by repeated presentation of
members of the category
- We have some solid simulation evidence for pattern learning
- In the long term neurons are added to CAs, leave CAs, and even
die
- CAs can also fractionate into new CAs (a concept becomes two concepts)
- Initially, the network has no CAs
- CAs develop in the network
- The end result is that the network has a large number of CAs competing
for neurons
- Furthermore, the CAs themselves are changing, breaking into new
CAs, and even dying out
