Re: With vector utility is one less likely to get stuck in local maxima?
- From: jonesrob@xxxxxxxxxxx
- Date: Fri, 30 Mar 2007 11:26:58 GMT
On Mar 27, 5:20 am, "Ted Dunning" <ted.dunn...@xxxxxxxxx> wrote:
On Mar 25, 6:33 am, "hegal...@xxxxxxxxx" <hegal...@xxxxxxxxx> wrote:
I donot agree that fitness has to be scalar. In fact, it is when
either you are evaluating single objectively or weighted vevctor of
multiple values. This is not true for most of the real world problems.
A more intuitive way is to design a multiple objective vector which
can be used to evaluate choromosomes. This is what multiobjective EC
about. The "probability of passing genetic information to later
generations" can be decided by the comparisons of all objectives in a
dominating/non-dominating way. Optimization doesnt have to be within
one-dimension also, i.e. in the case of single objective.
Probabilities are inherently scalar. That really isn't a matter of
debate. As such, evolutionary algorithms are computing the fixed
point of a Markov process
My claim was that utility might need to be a vector.
To quote von Neumann and Morgenstern (Theory of Games
and Economic Behavior, P.U.Press, 1944) "We have
conceded that one may doubt whether a person can always decide which
of two alternatives...he prefers... It leads to
what may be described as a many-dimensional vector concept
of utility."
I agree with von Neumann.
[ comp.ai is moderated ... your article may take a while to appear. ]
.
- References:
- Prev by Date: Re: With vector utility is one less likely to get stuck in local maxima?
- Next by Date: CFP: ECAL2007 deadline extension
- Previous by thread: Re: With vector utility is one less likely to get stuck in local maxima?
- Next by thread: conference on Artificial Intelligence and Pattern Recognition
- Index(es):
Relevant Pages
|
|
