From: Chris Capel (pdf23ds@gmail.com)
Date: Thu Feb 09 2006 - 12:59:16 MST
On 2/9/06, Richard Loosemore <rpwl@lightlink.com> wrote:
> Chris Capel wrote:
> > On 2/8/06, Jef Allbright <jef@jefallbright.net> wrote:
> >> Came across this lucid and current paper that touches upon much of the
> >> subject matter of this discussion list. Recommended.
> >>
> >> http://cscs.umich.edu/~crshalizi/bulletin/logic-of-diversity.html
> >
> > I wonder how the introduction of random steps, a bit of noise, into
> > the other heuristics of the strong heuristic agents would affect their
> > aggregate performance. I wonder if this was considered. My intuition
> > is that putting some noise into the search process (especially in a
> > simulated annealing sort of way) would be a big improvement.
> Interesting you should put it that way, because my first thought was
> that the role that "diversity" played *was*, in effect, equivalent (or
> at least, akin to) to simulated annealing.
Yes, that's a much better way of saying what I was thinking.
> The purpose of SA would be to keep the temperature up high enough to
> allow a global search for the largest-scale minimum, then lower the
> temperature and progressively look for the best of the smaller-scale
> minima within the larger one.
>
> Of course, the reason this would not be appropriate (I'm reversing
> myself now) is that SA has the fault that if there is an extremely good
> but very isolated (=unexpected, quirky) solution, it'll miss it.
I thought about this, but would simulated annealing be at a
*particular* disadvantage in that case? More particularly, I can't see
how it would be at a disadvantage compared to the naive heuristics of
the linked paper.
Chris Capel
-- "What is it like to be a bat? What is it like to bat a bee? What is it like to be a bee being batted? What is it like to be a batted bee?" -- The Mind's I (Hofstadter, Dennet)
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT