Copy Link
Add to Bookmark
Report
Alife Digest Number 013
Artificial Life Digest, Number 13
Thursday, April 5th 1990
Issue's Topics:
re: GAs and binary coding
re: GAs and binary coding
information-agents
----------------------------------------------------------------------
Date: Wed, 4 Apr 90 12:46:40 PDT
From: schraudo%cs@ucsd.edu (Nici Schraudolph)
Subject: re: GAs and binary coding
> From: ds1@philabs.philips.com (Dave Schaffer)
>
> IF strings are naturally ternary {0,1,#}, so we have the problem
> Christer Ericson mentions, what to do with the extra bit patterns?
> Smith used a redundant mapping:
>
> 00 -> 0
> 11 -> 1
> 01,10 -> #
>
> When my system had difficulty learning its task, I focused in
> on this redundancy. There are many different genotypes that
> represent the same phenotype. Furthermore, two parents with
> "#" at some position may produce offspring with either of the
> other alleles giving crossover a mutating effect. These bad
^^^
> effects had to be weighed against the richer schema sampling
It seems to me that in the context of classifier systems the side
effects of Smith's representation may be an advantage rather than a
handicap: since crossing a '0' with a '1' produces two "don't cares"
as offspring (and vice versa) this representation endows the crossover
operator with generalization and specialization capabilities. Could
it be that this, rather than the use of binary coding per se, caused
Smith's coding scheme to come out ahead of ternary coding?
Nici Schraudolph, C-014 nschraudolph@ucsd.edu
University of California, San Diego nschraudolph@ucsd.bitnet
La Jolla, CA 92093 ...!ucsd!nschraudolph
------------------------------
Date: Wed, 4 Apr 90 15:42:40 EDT
From: "David M. Chess" <CHESS@ibm.com>
Subject: re: GAs and binary coding
ds1@philabs.philips.com (Dave Schaffer) writes:
> The number of schemata instantiated in an individual is 2**L,
> where L is the string length. Clearly there are many more schemata
> instantiated in the individuals using the binary representation.
On the other hand, isn't it reasonable to look on the number of
schemata instantiated as 2**L, where L is the number of *bits*
in the string? So if I have a 4-ary encoding, the number of
schema instantiated is, in some sense, 2**(2*S), where S is
the string length. You get this by only a very slight
generalization from the usual Holland concepts, I think.
The usual concepts regard an individual:
0 1 0 0 1 1 0
as a "sample" from a large number of schemata, each of which has
one or more (well, zero or more) positition identical to the
individual. So this individual is a "sample" from the set of
individuals starting with "0", from the set of individuals ending
with "0", from the set whose second and fourth positions are "1" and
"0", and so on.
Isn't it just as reasonable to regard a 4-ary individual:
1 0 3 0 0 2 0
as being, as well as a sample of all individuals beginning with
"1" and so on, also a sample from the set of all individuals
whose first position is odd? I think that's just as valid
as regarding it as a sample from the set of individuals ending
"0 <something> 0".
That is, you can trivially rewrite any individual written as an
n-ary string as a binary string instead, by expanding each position
to its binary representation. So all representations really -are-
binary in some sense (modulo roundoff error due to non-power-of-two
arities).
This may all be absolute rubbish, of course! Rational replies
to that effect are most welcome...
DC
IBM T. J. Watson Research
------------------------------
Subject: information-agents
Date: Thu, 05 Apr 90 13:40:21 +0100
From: iaf%computer-lab.cambridge.ac.uk@nsfnet-relay.ac.uk
> I read an interesting article in a swedish computer magazine. It claimed
> that Nasa is investigating the possiblity of so called 'information-agents'.
> That is some kind of virus-alike program that should cruise the Internet
> in search of some specific information.
I recently attended a workshop on Belief Representation and Agent
Architectures at SRI-Cambridge (March 22-23). One of the presenters
- Phil Stenton (sps@com.hp.hpl.hplb) - talked about a system being
developed at HP Labs (Bristol, UK) called Wizard. It's basically a
very sophisticated user-interface package, providing such features as
speech recognition/synthesis, multi-media capabilities, network-wide
database querying, natural language report generation, etc, etc!! Every
user has an agent which acts on his/her behalf in order to reduce the
amount of information (potentially spanning many large networks) that
he/she needs to know about.
Anyway, it's "under development" and it is a commercial system, but he may
have written something about it. I suggest you contact him.
Innes A. Ferguson
Computer Laboratory, University of Cambridge, Pembroke St.,
Cambridge CB2 3QG, England.
BITNET: iaf@cl.cam.ac.uk
------------------------------
End of ALife Digest
********************************
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=---=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
= Artificial Life Distribution List =
= =
= All submissions for distribution to: alife@iuvax.cs.indiana.edu =
= All list subscriber additions, deletions, or administrative details to: =
= alife-request@iuvax.cs.indiana.edu =
= All software, tech reports to Alife depository through =
= anonymous ftp at iuvax.cs.indiana.edu in ~ftp/pub/alife =
= =
= List maintainers: Elisabeth Freeman, Eric Freeman, Marek Lugowski =
= Artificial Life Research Group, Indiana University =
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=---=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
End of Alife Digest
********************************
