In article <1993Apr3.032038.14001 at Princeton.EDU> gsanders at atbat.Princeton.EDU (Gillian Denise Sanders) writes:
>Hi. I am currently working on my undergraduate thesis on neural networks and
>mathematical models. I have been using the backpropagation algorithm and have
>noticed several references to its biological impausibility as one main
>disadvatage. I am taking a brain course and our professor mentioned retrograde
>messengers , nitric oxide and"back talking" in methods of learning such as
>classicalical conditioning. To me this would seem to suggest that perhaps the
>backpropagation method is not entirely biologically implausible...does anyone
>have any information about this topic or else any references to papers which
>discuss retrograde messengers and its backward synaptic directions? Thanks in
>advance. Gillian Sanders
Backprop involves an external *teacher*, whereas the weight change via
Nitric Oxide is a local phenomena more akin to a Hebbian type network.
Perhaps a more plausible backprop analogy might be found in the cerebellum
where an error signal is thought to originate from the climbing fibers
(inferior olivary nucleus).
1
UI - 93042981
AU - Chapman PF;Atkins CM;Allen MT;Haley JE;Steinmetz JE;
TI - Inhibition of nitric oxide synthesis impairs two different forms of
learning.
SO - Neuroreport 1992 Jul;3(7):567-70
2
UI - 92153413
AU - Haley JE;Wilcox GL;Chapman PF;
TI - The role of nitric oxide in hippocampal long-term potentiation.
SO - Neuron 1992 Feb;8(2):211-6
NEURON
I don't read the newsgroups too often, so if you would like to discuss
this further, please send e-mail!
Cheers!
-Rogene
*******************************************************************************
Rogene M. Eichler West
Neuroscience Program Minnesota Supercomputer Institute
University of Minnesota Army High Performance Computing Research Center
AHPCRC: (612)626-8082 1100 Washington Ave South
Home: (612)292-0573 Suite 101 MN Tech Center
Fax: (612)626-1596 MPLS, MN 55415
*******************************************************************************
