Re: Simulating Individual Behavior

[Mattias V. Bertelsen]

I have a question:  Why would a model need to be either ABM/IBM or
ODE/PDE?  It would seem that a model which even approached natural
complexity would be extremely difficult to construct or analyze if one
held strictly to one paradigm or the other.  I think that Swarm lends
itself quite handily to constructing models where the framework is
agent-based, but the behavior of the agents themselves can be modeled
using differential equations (for cases where the continuity assumption is
justified), statistical models, or explicit behavioral descriptions.

As soon as finals are done with, I am hoping to get started on my "summer
coding project": A Swarm object using multi-layer perceptron code (neural
networks) that functions as a black-box statistical model.  You give it an
input file of data and train a network to fit a statistical distribution. 
The object can then be used along the lines of the current distribution
objects in Swarm--hook up a pseudorandom number generator, and pull
numbers out of a distribution which came from real data.  I could see this
being used, for example, as a "weather object":  the intervals between
rain events in a simulated prairie ecosystem are reasonably modeled
without having to model butterflies flapping their wings.  <grin>

Any thoughts?

By the way--I am looking for a graduate program in CS or Ecology to work
with this stuff, starting around the fall of '98.  Any suggestions?

Mattias V. Bertelsen
address@hidden
http://www.spaceship.com/~mattias

On Fri, 18 Apr 1997, Mark P. Line wrote:
> Scott Christley wrote:
> > 
> > I am tending to notice a separation between two modelling paradigms: the
> > ABMers (I was an IBMer once! ;-) and the ODEers.  Are there true
> > differences between these two paradigms or is it only a perception?
> > Meaning is not a Swarm program a symbolic (mathematical) description of a
> > model, just not as concise as a differential equation?
> 
> In ODE-based and PDE-based models, the abstractions made are the ones
> that are imposed by the kind of symbolic manipulation (analytic
> solution) that is possible with simple ODE and very simple PDE systems.
> In other words: calculus tells us the function has to be continuous, so
> suddenly we are forced into an abstraction of our population of
> wildebeests such that population size is a real number and population
> growth is a continuous function. Biologically (or sociologically, or
> whatever), we tend to remain rather unconvinced that either one of these
> assumptions is particularly realistic. ODE's and PDE's were invented so
> that problems could be solved analytically. But few interesting problems
> that we'd model with these formalisms are soluable analytically anyway,
> so there's no longer any good reason to use them, and one very good
> reason not to use them: they force on us an abstraction that serves
> merely a by-gone purpose and which is usually not warranted in the
> biology (or sociology).
> 
> [As long as you're just doing some simple thumbnail models of biomass
> and energy balances and what-not, you might be safe up to a point just
> doing your ODE's, of course. I don't want to deny that.]
> 
> Now I can say what I think the answers are to the questions you pose
> above. 
> 
> Are there true differences between these two paradigms? Yes, because
> ABM's don't force abstractions on us that are artifacts of a method
> whose day is past and which are not desired otherwise in our models.
> 
> Is a Swarm program just a less concise description of a the same thing a
> differential equation describes? No, certainly not. A Swarm program does
> not normally describe the effects of fractional wildebeests pairing up
> and producing fractional young. It describes real, whole wildebeests
> pairing up and producing real, whole young. If anything, a Swarm model
> is _more_ concise. 



                  ==================================
   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.
                  ==================================