Article View: comp.ai.genetic
Article #58Re: GA versus Simulated Annealing
From: David M. Tate
Date: Wed, 27 Jan 1993 14:36
Date: Wed, 27 Jan 1993 14:36
44 lines
2601 bytes
2601 bytes
chang@pecan.cns.udel.edu (Liang-Wen Chang) said:
>
>The GA is insensitive to encoding due to the way of crossover operation.
>However, we also can say it IS sensitive to encoding once the encoding
>method is chosen.
This is certainly true.
>Because any encoding of numbers will have some
>bits is significant than other bits, because such inhomogeneous of
>bit encoding, the crossover will produce offsprings in the 'neighborhood'
>of the parents. The radius of the neighborhood is random and variable
>depending on where the crossover points are cut.
Just a note: not all encodings are bits, and not all breeding operators
are "crossover". That's just one special case, and not always desirable.
>I haven't figured
>out the statistic nature of such variable radius w.r.t. particular
>encoding. The recent 'uniform crossover' method is more of multi-bit
>mutation than crossover.
Not at all. Mutation is disruptive, replacing alleles from the parents with
alleles that do *not* occur in either parent. Uniform crossover is perfectly
conservative, recombining alleles from the parents. Just because it doesn't
use *contiguous* alleles from one parent doesn't make it "multi-bit mutation".
There's not necessarily anything sacred about the particular sequence the
genes are encoded in.
As I see it, the role of breeding (be it crossover or some other operator) is
to rearrange alleles you've got hanging around already, and the role of
mutation is to introduce (or re-introduce) alleles you'd *like* to have around,
but don't. Viewed this way, uniform crossover is clearly unlike mutation,
and clearly like crossover.
Again, most of my work is with combinatorial optimization problems for which
bit-string encodings would be nightmarish. So my perspective is perhaps a
trifle unusual.
--
David Tate |"Something there is that doesn't love a wall,
dtate@unix.cis.pitt.edu | That sends the screaming liner over it
Prof. of Story Problems | And spills the bleacher-dwellers from their seats."
Member ORSA, TIMS, SABR | "Fenway Wall", Robert Frost
Technical Details
Message-ID:
<2624@blue.cis.pitt.edu>Path:
arc1-ca.pugleaf.net!sparky!uunet!ulowell!m2c!bu.edu!rpi!gatech!pitt.edu!geneticReferences:
<1993Jan19.155159.16736@thunder.mcrcim.mcgill.edu> <2360@blue.cis.pitt.edu> <1993Jan27.082732.15090@udel.edu>Show Full Headers
Complete Article Headers
Path: arc1-ca.pugleaf.net!sparky!uunet!ulowell!m2c!bu.edu!rpi!gatech!pitt.edu!genetic From: genetic+@pitt.edu (David M. Tate) Newsgroups: comp.ai.genetic Subject: Re: GA versus Simulated Annealing Message-ID: <2624@blue.cis.pitt.edu> Date: 27 Jan 93 14:36:37 GMT References: <1993Jan19.155159.16736@thunder.mcrcim.mcgill.edu> <2360@blue.cis.pitt.edu> <1993Jan27.082732.15090@udel.edu> Sender: news+@pitt.edu Organization: Department of Industrial Engineering Lines: 44