home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pipex!demon!cix.compulink.co.uk!mark_a
- Newsgroups: comp.theory.cell-automata
- From: mark_a@cix.compulink.co.uk (Mark Atkinson)
- Subject: k
- Cc: mark_a@cix.compulink.co.uk
- Reply-To: mark_a@cix.compulink.co.uk
- Date: Thu, 7 Jan 1993 18:10:00 +0000
- Message-ID: <memo.847952@cix.compulink.co.uk>
- Sender: usenet@demon.co.uk
- Lines: 44
-
- The evolving CA idea is very interesting....
-
- One thing occurs to me about Packard's "Disneyworld" proposal - 2D CAs have a
- very strong propensity to either die out of descend into chaos (of the
- uninteresting type). As Langton et al. have shown, the "best" (here,
- aesthetic) rules are balanced procariously on the "edge of chaos".
-
- How are we proposing to constrict the ruleset(s) such that our evolving
- pattern does not die? Is the correspondence between aesthetics and balanced
- entropy sufficient to ensure the observer selection-driven evolution will
- ensure our pattern will not go too far off-track?
-
- Perhaps using a class of CA which is less prone to these possiblities would
- be the answer. Any usual 8SUM (Toffoli's terminology) rule (even a very
- well-balanced one, like Conway's Life) would tend to stabilise after some
- time due the inevitable increase or decrease in its entropy level. I would
- propose using a time-reversible rule to avoid these problems.
-
- On the side of the genetic algorithm (GA), careful attention should be paid
- to the schema; that is to say, no small number of selections (which could
- potentially be contrived or even malicious here) should cause the pattern to
- diverge drastically from its current state. The evolution must proceed in
- very small steps if it is to achieve good results, but no so small as the
- users to not see feedback within their short attention span. There is a
- difficult dilemma here.
-
-
- On the "alternatives" front, IFS codes (Iterated Function Systems) are a good
- bet - aesthetically pleasing, small codes giving rise to complex structures,
- robust/similar under mutation/recombination. On the minus side, they use
- real numbers, but this isn't too much of an issue.
-
- I have written a "Blind Watchmaker" type selector for IFS patterns, and the
- results _are_ nice.
-
- =============================================================================
- Mark Atkinson
- mark_a@cix.compulink.co.uk
- -----------------------------------------------------------------------------
- "This statement is true, but you cannot prove it." - Kurt Godel.
- My opinions are shareware - if you like them, please fax me $25.
- =============================================================================
-
- >>>MATRIX version 1.21c
-