Q about artificial and biological neural-nets

lamontg at u.washington.edu lamontg at u.washington.edu
Sat Feb 27 18:12:54 EST 1993


not sure if this is the best forum, but...

recently, i was having a disucssion with a friend of mine over a 
hypothetical model i have of a small amount of "neurotoxicity" effecting
specifically axons, possibly being actually functional in terms of
producing a more efficient, and well-adjusted neural net.  the hypothesis
runs something along the lines of that the destruction of the connections
between the neurons promotes the growth of new connections, and that the
connections that are not re-established could be seen as being relics of
a previous state that the neural net was in, while the ones that are
re-established could been seen as being only the ones that are 
applicable to the environment that the neural net is in now.  and the
net result is that it functions better now than before.

then, he mentioned something which had been shown along these lines in
artificial neural nets.  it was something along the lines of if you
trained the neural net to behave in one way, that simply trying to
retrain it to behave another way wasn't as effective as first "damaging
it" (whatever that means) and then retraining it.

so, i'm looking for clarification on what exactly it is i'm taking about,
and what kinds of thoughts are already out there on this topic...
 
-- 
"Of all the strange 'crimes' that human beings have legislated out of 
nothing, 'blasphemy' is the most amazing--with 'obscenity' and 'indecent
exposure' fighting it out for second and third place."
                                    -- Lazarus Long



More information about the Neur-sci mailing list