Topology
- Multi-Layer Perceptrons are not biologically plausible.
- The brain is sparsely connected.
- Hopfield is out.
- We use toroidal topologies to avoid border effects (thanks Paul).
- Sets of interconnected toruses are the default, but in the CABots
we often engineered the topology.
- In general we used distance biased connectivity for excitatory neurons.
- We use random connectivity for inhibition.
- In the brain if there are connections from one area, there are
connections back.
- I used small world topologies to get the right kind of persistence.
Things should persist like STMs, relatively briefly (a few hundred
cycles, a few seconds).
