From: Rüdiger Koch (rkoch@rkoch.org)
Date: Mon Apr 01 2002 - 23:52:49 MST
You got me pants down ;-)
A rough idea is:
dream up a simple function for jam-smell-molecule concentration in the air,
maybe:
smell = k * 1/(distance^2)
then take the relevant input neurons and determine how to let them get
excited. Just how that happens I don't know yet because I don't know how and
how long a molecule excites a receptor neuron. I guess I'll have to figure
that out for C.Elegans ASAP since a few hours ago I commited to Eugene Leitl
for this challenge.
The sentence: 'this is jam' -> 'jam **activates** neurons 34 to 102' doesn't
make too much sense in an Amygdala network. Amygdala has neuron IDs, of
course. An Amygdala neuron gets spike input from other neurons, the
**timing** of which determines whether it will fire or not (sending out a
spike). I don't know if this integrate and fire type of neurons is good
enough a model (Penrose says NO), but I think it is. Your sentence make sense
if you think about spike rate neurons. This is definitelly not enough because
spike rate is not a very important measure (except for motor neurons and
such). There is no way to get a C.Elegans simulation with SNNS or PDP++.
-Rudiger
On Tuesday 02 April 2002 07:02, Alejandro Dubrovsky wrote:
> On Tue, 2002-04-02 at 07:12, Rüdiger Koch wrote:
> > Hey, words are cheap. Where is the code? -> I've heard that it's custom
> > to offer bets on this list. What about this: If someone gives me a map of
> > a drosophilia brain with all neurons and their connections we'll have an
> > Amygdala NN that likes to eat virtual fruit and fly around virtual jam
> > jars in less than 8 months. USD5,000. , anyone?
>
> The slides for C. Elegans are there. Much smaller and simpler, and i
> think there's almost no way it will be done in under 8 months. (Even
> the physics simulation of the external world and body in the fly's case
> would take much, much longer than that. How is the virtual fly meant to
> recognise the virtual jam for what it is? are you suggesting hardcoding
> a symbolic 'this is jam' -> 'jam activates neurons 34 to 102' in which
> case jam-analogues that would excite the real fly would not excite the
> virtual fly, or are you going to simulate jam's molecular chemistry and
> fly's receptor, in which case your computational needs just jumped by a
> fair bit i imagine. Even in the C.Elegans case, i think this would be
> harder than the neuronal simulation)
>
> > There is no need to simulate on the molecular level - capturing all
> > relevant information theoretical aspects of spiking and synapse strength
> > adjustments, maybe also gap junktions (and leaving out all quantum
> > information theoretical micro tubule bull^H^H^H^H theories) will get us
> > there, I am sure of that.
>
> agreed, FWIW, but only inside the brain. for the external world, i'm
> not so sure.
> Alejandro
-- Rüdiger Koch http://rkoch.org Mobile: +49-179-1101561
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:13:11 MST