Overlapping Cell Assemblies
- Most modelling involves non-overlapping CAs. That
is, a neuron is in at most one CA.
- Sakurai has pointed out the importance of overlapping CAs.
- Firstly it enables more CAs to be put into a given network.
- Secondly, and perhaps more importantly, it allows different
CAs for related concepts.
- To get this to work at all, we've needed Compensatory Learning,
which uses the total synaptic strength of the neuron as part of
the learning rule. (We also used it in the spontaneous neural
activation work.)
- This is work that has just come out in Neural Computing.
- We took our standard 20x20 net and started to include
more and more rows in both patterns.
- This shows that up until about 40% overlap we get
two distinct patterns.
