Alt-BEAM Archive
Message #00744
To: Sean Rigter rigter@cafe.net, beam beam@corp.sgi.com
From: Noam Rudnick rudnick1@cwix.com
Date: Sun, 21 Feb 1999 16:37:08 -0500
Subject: [alt-beam] Re: associative memories
Good question. I was thinking more like visual memories. For instance, if
you happen to be rumaging through you toolbox(and a messy one at that!),
looking for a screwdriver and you see the handle of one, you can associate
that handle with the screwdriver, and pull it out. Now although we take
this type of behavior for granted, robots don't. Most traditional cpu-type
robots need the full image. Not anly that, but there cannot be any visual
noise. The advantage of an associative memory is that you (humans) are
able to associate the handle with the rest of the tool. This is only one
example.
Now let me rephrase my question (in relation to the above example):
Say I also wanted my robot to remember a wrench, in addition the the
screwdriver. is this just a matter of adjusting the resistances between
each neuron? At what point does it become impossible to hold any more
memories? Never? Or will the robots ability to differentiate between each
tool go down, as the number of tools increases?
Another way in which associative memories could be used would be to
associate ideas (in a sense). For instance, say a robot encounters another
robot which is making loud noises, and moving quickly. Then the other
robot drains all the power off of the original one. Lucky for him, the
original one has backup solar power :) so he comes back to life. This time
though, when he encounters a fast moving robot that makes a lot of noise,
it will run away as fast as his little legs can carry him, because he
associated loosing all his power with that type of robot.
I hope these examples cleared up any misundersatndings about my original
question.
Thanks,
Noam Rudnick
----------
> From: Sean Rigter
> To: Noam Rudnick ; beam
> Subject: Re: associative memories
> Date: Sunday, February 21, 1999 3:49 PM
>
> Hi Noam,
>
> It's Wilf , Sean is my oldest (1 of 5) offspring who is kind enough to
> let me use his fast internet access computer while I'm home.
>
> The question is : what is the question? Memory of what?
> Do you mean memory of a sequence of steps? Or a sequence of external
> inputs? Or an internal landscape representing a history of collisions vs
> steps. Can't be deterministic linear sequence so needs pattern
> recognition (associative memory) The walker can traverse it's internal
> landscape,(acting out it's memory) sampled alternately with the real
> world, and ?
>
> The latter may be interesting since it would provide a learning mode.
>
> Noam Rudnick wrote:
> >
> > Hello Wilf or Sean, by the way, which one is it?
> >
> > Your LIFE ideas sound interesting. I was more interested in the memory
> > aspects of content addressable memory/associative neural nets.
> > Specifically, do you have any ideas on how memories could be stored? I
can
> > only think of hardwiring them, which feels to unnatural. Also, I know
it
> > would be possible to store at least one desired memory, but would it be
> > possible to change other stable memories without changing the first
one?
> >
> > Thanks,
> > Noam
> >
> > P.S. Maybe sometime we (and anyone else thats intersted) should go
into
> > the beam chatroom and actually talk about these ideas instead of
sending
> > e-mails back and forth.
> > ----------
> > > From: Sean Rigter
> > > To: Noam Rudnick ; beam
> > > Subject: Re: associative memories
> > > Date: Sunday, February 21, 1999 12:00 AM
> > >
> > >
> > > > Noam Rudnick wrote:
> > > >
> > > > oops, i guess i jumped the gun.
> > >
> > > No, I was just slow!
> > >
> > > In fact, if in your circuit you use 74HC14 gates with 2 identical
> > > resistors connected together at one end and connected to the input of
> > > each gate it forms 2 input "majority logic gates". If the voltage at
the
> > > other end of each of the 2 resistors is connected to input voltages
the
> > > output changes only if the majority of inputs is either one or zero.
ie
> > > 0V at both inputs change the output to 1 If both inputs are 1 the
output
> > > is 0 if there is no majority of one or zeroes at the 2 inputs, the
> > > output remains unchanged.
> > >
> > > Actually one correction (my fingers are faster than my brain!)
> > > I should have said the minimum neighborhood size for LIFE is 9 cells
> > > with 8 surrounding the central cell. In large arrays each cell is
> > > surrounded by 8 neighbors except at the edge of the array.
> > >
> > > XXX
> > > X0X
> > > XXX
> > >
> > > With non inverting majority logic gates the CA behaviour will be one
> > > called "VOTE". Cells turn on if a majority of neighboring cells are
on
> > > or turn off when the majority are off. If there is no clear majority
the
> > > cell output remains unchanged.
> > >
> > > Such a rule could also be used to control group behaviour of robots
if
> > > they are wired up with majority logic rules and can communicate with
> > > each other.
> > >
> > > Talk about peer pressure and conformity
> > >
> > > Anyway I wrote a short article on this which you can find on my
> > > absolutely unfinished web page
> > >
> > > http://www.geocities.com/ResearchTriangle/System/8222/beamlife.html
>
> --
> Hey, come visit my homepage at
> http://www.geocities.com/Area51/Nebula/1710/
> Southpark, Mechwarrior2, V&V, Al-Qadim, and Milla Jovovich!
------------------------------------------------------------------------
eGroup home: http://www.eGroups.com/list/alt-beam
Free Web-based e-mail groups by eGroups.com
Home