Learning
- The work described in these pages is about constructing a network
- The real work in CAs is about learning CAs
- The brain does not have the luxury of already knowing synaptic
structure.
- (My personal guess is that brain area structure is hereditary, and
anything below that is quite random and learned.)
- Consequently, the real question from all of this is how can these
types of networks be learned?
- A novel point from the Hopfield/CA work is that inhibitory neurons
need only inhibit nearby neurons that are not in an overlapping CA
- This may affect the topology of inhibitory neurons (e.g. distance
biased)
- We are currently calling our inhibitory neurons Chandelier Neurons, but
we could use Purkinje additionally or in place of
- Our model of actual neurons is at best a metaphor
- It may however influence the learning rule so that inhibition
is more intimately related to recruitment and
fractionation
- This may also relate to Mexican Hat like learning functions
