fLIF neurons and CAs
- I have been working with Cell Assemblies (CAs) and fatiguing
Leaky Integrate and Fire (fLIF) Neurons for about 8 years
now.
- I've given quite a few talks here on them, and I think they
are Turing complete.
- our fLIF neurons:
- integrate activity from attached neurons
- ours spike if they pass a threshold (sending activity)
- activity decays (or leaks) if the neuron doesn't spike and loses
all activity if it does spike
- neurons fatigue when they fire (threshold is raised), and
fatigue is reduced when they don't spike
- learning is done by a type of Hebbian learning
- connections are uni-directional
- individual neurons are inhibitory or excitatory but not both
- CAs are attractor states and groups of neurons that are highly
connected.
- CAs persist after input has ceased, but can be shut off.
- CAs are the basis of concepts.
- We've done a lot of work with categorisation.
- More recently, we've done work with variable binding and if-then
rules (which fits in nicely with NL grammars).
- We have a version of the simulator (CANT) that you can use
on the net, but if you want to try some simulations out, it's probably
best to get the current version.