Re: What are the advantages / disadvantages of Nueral Networks vs. Complex Adaptive-Agent based modelling?

[glen e. p. ropella]

At 03:42 PM 2/4/99 -0700, you wrote: 
>
> I'm doing the programming for a project that attempts to simulate "how
people
> choose" in order to simulate a small economy with lots of individuals, a
> couple of businesses, and a government that regulates.  I've read most of
the
> Sugarscape book and have a limited but good-enough understanding of the
> complex adaptive-agent technique.
>  
> How does this technique compare to nueral networks?  My impression of nueral
> networks is that they're much more complicated.  Does it even make sense to
> incorporate nueral networks into agent-based simulations?  I'm not even sure
> what they use Nueral Nets to do, anyway.
>  
> If anyone has a minute to spare, please enlighten me!  :)


Hey Matt!

ANNs are primarily useful for pattern recognition, associative memory,
and controls.  The fundamental difference between an ANN and an ABM
is that of abstraction.  First off, they're both "networks".  An ABM is 
a network with an indefinite set of node and link types (i.e. the computation
inside the node can be anything and the signal that runs along the links can
be anything).  An ANN is a specific kind of network, namely one that works
on the basis of an activation function.

So, for considering things like Swarm, ANNs are extremely useful as an
agent's internal processing.  But, they are inapplicable to most of the 
types of problems where the network is inhomogenous as to node or link
type.  And using Swarm to implement an ANN would simply be crazy, even
though it could be done.

Does that help?

glen


--
glen e. p. ropella                =><=          Hail Eris!
Home: http://www.trail.com/~gepr/home.html  (505) 424-0448
Work: http://www.swarm.com                  (505) 995-0818  


                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================