The Cell Assembly Mark 3 (CA3)
- Most CA work has focused on roughly binary CAs.
- The attractor net work just uses a Hebbian calculation.
- Most of our stuff has had CAs that are really just on or off.
- There can be some priming.
- There can be some sub-ignition firing.
- However psychological memories can be more or less active, not
just on or off.
- This is important for our Maes net, but also for associative
memories, parsing and psychological modelling.
- So, following Milner's 1957 work on the CA mark 2 (mark 1 is Hebb's
original), we propose to look at the CA mark3.
- The problem is still ill posed, but other issues we're considering
include resolving the stability plasticity dilemma, short-term memories
fading correctly, and associative memory.
- We'd like to use known neural properties to get more
psychological properties.
- We'd then like to fold this medium term work back into the CABot
agents.
