Conditioning?

Harry Erwin erwin at trwacs.fp.trw.com
Sat Oct 23 18:30:59 EST 1993


In article <1993Oct23.013721.5873 at leland.stanford.edu> 
dwheeler at leland.Stanford.EDU (David Brockman Wheeler) writes:
>>>... In other words, LTP does not actually
>>This is hardly the thing a neural network programmer wants to hear. 8)
>How come?
It implies that things are a great deal more complicated than I was
hoping. Walt Freeman told me that I had to understand habituation and
conditioning in some detail to be able to put together a trainable version
of his KII/KIII models of the olfactory system. Now it appears some of
those details are simply unknown. I had had a strong suspicion that this
problem would eventually provide me with a dissertation topic, but it
looks like it's much harder than that.
>>...{possible}trim it down on one side and pump it up on the other.
>The current model of LTP has the induction occurring postsynaptically whereas
>maintenance (expression over hours or days) involves both pre- and
>postsynaptic changes. Thus, I am not sure how we could describe 
>a connection such as that outlined above. I would like to point out, however,
>that synaptic plasticity is not an all-or-none event, ie there is a continuum
>from LTD to LTP and I would not be surprised if a period of sub-threshold
>stimulation (ie tetanus insufficient to induce LTP) could have opposite
>effects on either side of the synapse. Just warbeling (sp?) but would be
>really interested in knowing how or why this might be useful d:^}

If LTP is induced post-synaptically, then I might expect that stimuli of
the right type and magnitude so that the post-synaptic neuron does not
fire would result in short-term habituation on the pre-synaptic side and
short-term potentiation on the post-synaptic side. Then, if the system is
sensitized, so that the habituation is lost, a follow-on signal might
really blast through despite having been shunted to ground the previous
time through. Modelling a training process involving signals that are
discarded in this way might be quite interesting, for a couple of reasons:
1. The system would be functioning as a long-term integrator, and
2. I can see where there would be some impressive non-linearities in the
dynamics.
Inhibitory axosomatic connections might work with this system to make it
more important than might be obvious at first glance. The term
"hair-trigger" occurs to me. 

I think I have to go off and think about this one. I've seen a couple of
"hair-trigger" processes in sensory systems that this might explain.

Cheers,
-- 
Harry Erwin
Internet: erwin at trwacs.fp.trw.com
          herwin at cs.gmu.edu
Working on Freeman nets....



More information about the Neur-sci mailing list