Processing with fLIF Neurons

You'll note that CAs are all about stable states.
This is great for attractor nets (e.g. Hopfield), but it's lousy for cognition.
That is, you have to be able to move onto new states.
We thought we could do this, but have now actually done it.
Sequence is done by having CA1 excite CA2, and CA2 inhibit CA1.
We've got a little bit of code that makes this all flexible on decay rate, so that you can get a range of processing speeds.
Conditional changes of state can be done by having CA3 only come on when CA1 and CA2 comes on. Lets do neuron to neuron connectivity with reverse inhibition. D=2 and theta=4. If the forward connections are 1.5 then CA2 fires up iff CA1 and CA2 are up.
We did this with "CAs" of 6 neurons, and showed that all FSAs could be built with fLIF neurons.
These things are Turing complete. (I implemented a stack, so could implement 2 stacks, + FSA makes Turing complete.)
That's not too surprising. The continuous version is super-Turing complete (Siegelman I don't understand it.)
So, you can process with fLIF neurons. The question is now how do you learn to process?