Weighting of synaptic responses

Rose Reich rreich at shell02.ozemail.com.au
Tue Aug 27 17:01:31 EST 1996

Matt Jones (jonesmat at ohsu.edu) wrote:
: In article <y4bufxx7uc.fsf at mailhost.neuroinformatik.ruhr-uni-bochum.de>
: Jan Vorbrueggen, jan at mailhost.neuroinformatik.ruhr-uni-bochum.de writes:

: >Define "standard neural net modelling".

: Well, I'm not really a connoisseur of neural net modelling, so maybe
: _you_ could provide a brief description of the most common paradigms in
: use. That would be helpful. And do any of them include inhibitory
: synaptic connections (not negative alteration of weights, but active
: inhibition)?

A standard feedforward neural net breaks 3 neuroscience principles:
a) No Loops
b) No ISPS
c) No frequency encoded APs

But this is really irrelevant to my question as 
a) If i want to break any of these rules, I can. The only effect is to make
my simulation more accurate/complex.
b) I am only considering an individual neuron, whose properties i will 
discuss below :

A neuron (computationally) is made up of 3 main units, with physiological 
parallels :

Dendrites: Receive inputs from other neurons (usually an analog signal, 
as they are much easier to deal with than All or None AP's)

Weight Units: Multiply the dendritic input by a factor that can be 
modified by the controlling program. Eg if a single dendritic input is 
100 and the associated weight of that synaps is 0.8, then the value sent 
to the hilock is 80.

Hilock/ Integrating unit: This sums all of the inputs and passes them 
through a sigmoid squashing funtion. And the output is available for 
other neurons to use as inputs.. etc...

I hope this helps to clear up my question. And thanks for all of the 
replies i have received so far.


More information about the Neur-sci mailing list