Re: Intelligent ways to search parameter space?

[Stephen C. Upton]

Hi Steve,

I believe you're right on the idea.  Also, that's what I meant, and I
think that's what Miller had in mind, is that even though you formulate
the problem as an optimization one, doesn't mean you have to look at any
specific parameter, and the goal may, in fact, be to characterize the
landscape.  For example, say you're interested in regions of the
landscape that are chaotic, then you could construct a fitness function 
(not sure how to do this yet!!) to be a maximum for these regions, and
then the search routine would preferentially find those points in
simulation space that would be chaotic.  You could then analyze these
sets of points to see what "caused" the chaos.  I guess it's a
roundabout way of getting at the inverse problem, i.e., given a y
(chaotic region) finding the x.  Also, in my case, I'm also interested
in relatively "flat" regions, as those represent "robust" conditions for
my domain.

I'd also be interested in a copy of your dissertation when it comes off
the "presses".

thanx
steve

Steve Shervais wrote:
> 
> Steve,
> Must be an idea whose time has come. I used GA optimization to search
> parameter space in an inventory control problem for my dissertation --
> "in press" as they say. The goal was not so much optimization of any
> parameter as characterization of the search space.  The idea was that a
> GA would spend more time near the "hilltops". These are areas where you
> want your controller to really understand the problem in detail. The GA
> will spend proportionally less time in the "flats", where what you need
> to know is how to find the nearest hill. Turns out, on the problem I
> addressed, there was no difference between random, grid-stepping, and GA
> parameter selection, but that might just be the fault of my specific
> problem (very spiky, deceptive landscape).
> 
> Steve Shervais
> Assistant Professor, MIS
> Eastern Washington University
> address@hidden
> 
> "Stephen C. Upton" wrote:
> >
> > As somewhat of an aside, I am also finishing up a software framework for
> > using Natural Algorithms (GA's, simulated annealing, ant algorithms,
> > cultural algorithms) to "search" over simulation spaces. (this is part
> > of my long overdue dissertation).  The basic idea is to formulate your
> > search as an optimization problem and then use the Natural Algorithms as
> > heuristics to find the "minimum" or the set of minimum.  This is the
> 
>                   ==================================
>    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.
>                   ==================================



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