X-Message-Number: 27546
Date: Sun, 29 Jan 2006 11:34:10 EST
Subject: Uploading technology (1.v.2) Shortcomings. 

Uploading technology (1.v.2) Shortcomings.

In a project such  this one, that is producing a uploading quality neural 
network, there are a  number of shortcomings. The first comes from the ongoing 
research process: New  informations are published everyweek or month, it is 

impossible to wait unitl  everything is known and analysed. At some time we must

say stop and take the  body of knowledges at this instant. This will define the
modeled neuron such it  is understood at a particular time. My feeling is 

that what is known is largely  sufficient to support an interesting electronics

Another  shortcoming come from my limited knowledge, I have not read 

everything published  on the subject in the past thirty or forty years. For 
example I 
am somewhat  foggy about the thresholds of the one thousand or so g-protein 
gated channels.  My feeling is that a large part of that diversity has a 

biological interest, for  example for surviving some illness but there is no 
information meaning  here. A set of arbitrairy thresholds, every 5 mV for 
example would fit all the  information processing needs. In the same way, a 

detailled knowledge of each  neurotransmitter may not be important in a first 
if conduction,  thresholds, duration, maximum potential and inactivation are 
taken into account,  there is hardly another parameter able to distinguish  a 
current from a  neurotransmitter from another.

There are too some intended  shortcomings, for example excitatory synapses on 
spines will model a number of  spine properties, this may go beyond the 
capacities of real spines. On the other  hand, dendrite domain are not so well 
served, it is impossible to alter a  defined dendrite branching without 

reinitialisation, this is linked to the  limited number of I/O pins on current 

Many parameter starting  values will be defined "by hand". The corresponding 
biological element being  nearly impossible to mesure. One way out of that 
situation would be to have a  brain reader or at least a neurons reader at 

molecular level and then run a  simulation on a computer from the first quantum

mechanics principles. This would  define the parameters and link them to what is
seen by the brain  reader.

Another important limit is that current FPGAs may contain  only someting as 
300 neurons, this is two or three microcolumns, the smallest  brain unit after 
the neuron. This technology could go to the column level if  more chips are 
used and real time speed is not a constrain.

These  shortcomings and other come from something found in all technological 
projects:  There must be some trade-off between some parameters. Some choices 
must be made.  If we don't accept that and ask only for the perfect system, 
nothing will be  never done.

Yvan Bozzonetti.

 Content-Type: text/html; charset="US-ASCII"


Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=27546