I've just done some work on possible neuropeptide transmission in
neural nets (for my PhD). Despite some of the rather shaky assumptions
I've built in, and the inherent shakiness of NN plausibility in modelling
real neural processes (I used back-prop), I got some enhancement in
learning speed.
The basic idea was that peptides could act as a 'broadcast'
signalling mechanism, such that they would diffuse from a source cell,
and represent a modulation (on a very small scale) at receptors. The
enhancement in learning speed compares roughly (I like to think 'is a
little better than') random noise injection at its best.
Of course, this ignores questions like 'Don't glia get in
the way?' and 'Is signal weighting pre- or post-synaptic?' (I assumed
pre-). I think more research needs to be done on this, but I couldn't
get funded to do it. (Even though I got offered a post-doc at Caltech)
If there's anyone out there who wants to employ me to do this
research, it would make me very happy.
-Phil Docking.