Attractor States
- CAs are attractors, but from feature neurons
- Similar patterns activate the same CA
- Of course these final states are groups of neurons, and there can
be some variance
- There should be a large amount of overlap (Pearson value)
- Given the environmental stimuli, the net categorises that stimuli
by moving to an Attractor State (CA) that has many neurons
firing.
- We want CAs for a variety of categories.
- Some closely overlapping, some sparse
- How can we form CAs for all of these?
- (We've assumed, tacitly until now, that the input neurons are part
of the CA. This is probably not the case with sensory neurons.
There are probably some cross brain area dynamics that we really
do not know much about now. (e.g. LGN<->V1))
- How does a layered topology affect the problem
