Walter Eric Johnson
wej3715 at fox.tamu.edu
Wed Dec 9 08:20:39 EST 1998
david_olmsted at my-dejanews.com wrote:
: In article <74fr9l$l4l$2 at news.tamu.edu>,
: wej3715 at fox.tamu.edu (Walter Eric Johnson) wrote:
: > david_olmsted at my-dejanews.com wrote:
: > : For a negative view to that idea see my site at
: > : The alternative to the more common idea that neurons process data using
: > : spatio-temporal integration is that neurons process data using
: > How about neural assemblies?
: > Eric Johnson
: I assume by neural assemblies you mean the work that has been done with
: neural network theory. I am just not impressed with their ability to process
: information in an efficient and biologically realistic fashion. Learning
: takes too many trials and beyond simple pattern classification they can't do
: much else. Also these neural assemblies must still assume some basic neuron
: operations which in the past has always been the standard neuron model with
: the threshold.
"Neural network theory" = "Artificial Neural Networks" ? When you
mention the too many trials and simple pattern classification, I
think you are discussing ANNs, but then the next sentence makes me
doubt it. So I'm not sure what you meant.
Anyway, in neural assemblies, the important thing seems to be that
individual neurons may form functional groups for processing and
the emphasis shifts from the actions of one neuron to the actions
of populations of neurons working together. Thus, the population
encoding of the neural assembly would predominate over the
output of any individual neuron.
It seems to me that the rationale behind neural assemblies is
the opposite of the rationale behind microcircuits as you advocate.
(Somewhere I saw the word microcircuits applied to neural
assemblies, but I really didn't understand why they used that
particular term in that instance.)
More information about the Neur-sci