Current State
- Finite State Automata: I've got a class with states persisting at
5 ms. They
can persist faster (3 or 4 ms) or slower (up to 9 ms), but behaviour
becomes unclear.
- There are several tests. (They don't run in a suite.)
- I use it for parsing, cognitive mapping and a bit of the vision.
- Writing the cognitive mapping, I can see how the FSA Class could
be improved.
- I'm hoping that a change in neural, synapse or input could be largely
fixed by changing the FSA Class and passing the tests.
- Similarly, I'm hoping this will also work for HiCANN.
- Bug 1: found a crash on more the 255 synapses in PyNN 8.0 with Nest 2.6.
Andrew D. resolved, in PyNN 8.1 with Nest 2.10. Testing that in
docker shell, and on Ian' machine.
Bug 2: IF_cond_exp neuron fires if inhibitted too strenuously in
SpiNNaker. Work around is to use FSAs for inhibition with states
persistently inhibitting.
- CABot3 in SpiNNaker with virutal environment. It has
NLP, vision, and a planning stub.
- It runs in Nest but the environment is not plugged in for vision.
- Cognitive mapping module binds the rooms to the shapes in
SpiNNaker (and almost in Nest).
- Planning is undergoing a rewrite.
- Papers at PPIGs (neural FSAs as a software design pattern), and
BICA (neural planning with Maes nets.
