Alt-BEAM Archive
Message #02962
To: "BEAM Mail" beam@corp.sgi.com
From: "Nigel Joyce" nigel.joyce@qr.com.au
Date: Wed, 5 May 1999 14:29:52 +1000
Subject: [alt-beam] Re: Multiple Inputs for Nv neurons
Sounds like it is fast becoming like the Classic Programming NP problem, The
NP Problem is a problem of finding the fewest/fastest/easiest/optimum
solution to a complex network of interconnected nodes.
Blue Skies
---Buzz---
nigel.joyce@usa.net
> When making more complex Nv/Nu nets, things get a lot tougher. You have
> to consider saturation though. If they are not tuned properly, a few
> bits of stimuli will put the thing into saturation easily. I believe
> Tilden implied that Lobster suffered from this problem in a message he
> sent a while back. If you do make large networks with interconnecting
> "web like" loops, then be prepared to tune the thing, and be prepared to
> spend time on it. A good summer project??? Maybe??? I'd like to see some
> new Nv setups.
>
> George Rix wrote:
> >
> > > Also, they need a form of regenerative feedback, or all the processes
> > > simply end. The loop does that. THe thing that you may miss in a web
> > > like Nv, is that there are many, multiple loops made up by all the
webs.
> > > You couldn't begin to count the massive loops in the human brain!!!
> > > That's quite the daunting task to do in even an earth work or
cockroach
> > > even! Loops seem to be the best Central Pattern Generator,a nd they
can
> > > branch off into chains, rejoin, form other loops and do countless
> > > amazing things. Unfortunately, BEAM is in it's infancy, and neural
> > > network based robots are well less than 100 years old (despite the
fact
> > > that the neural net was essentialy one of the first "control" devices,
> > > created by nature itself!). Neural nets have had a while to develop,
and
> > > we are only trying to begin to catch up!
> > What I'm planning on doing is having the neurons arranged in an upside
down
> > triangle, with the one at the bottom point being the first to go. From
there
> > they expand, with some interesting interconnections I've been planning
> > (okay, anyone could do them, but they're still interesting) and at the
end,
> > loop back to number one. THAT's what I meant.
> > In the twelve neuron robot I'm planning, there should be some48 loops (I
> > think, though I'm not sure. There are likely many more, but I got bored
of
> > counting =)
> > Signing off,
> > Rob Rix
> >
___________________________________________________________________________
> > Newsflash! A scientific report has just come in from Montreal stating
that
> > humans are unpredictable and volatile. If you spot them contact the
local
> > authorities. Do not attempt to approach them, as they may blame you for
> > making crop circles and fire high-speed explosive shells at you. Take
care!
>
> --
>
>
> Richard Piotter
> richfile@rconnect.com
>
> The Richfiles Robotics & TI web page:
> http://richfiles.calc.org
>
> For the BEAM Robotics list:
> BEAM Robotics Tek FAQ
> http://people.ne.mediaone.net/bushbo/beam/FAQ.html
------------------------------------------------------------------------
eGroup home: http://www.eGroups.com/group/alt-beam
http://www.eGroups.com
- Simplifying group communications
Home