Re: No Free Lunch (NFL) Lecture

From: bivalve (bivalve@mail.davidson.alumlink.com)
Date: Fri Jul 26 2002 - 19:57:47 EDT

  • Next message: Stuart d Kirkley: "Re: deception in perception"

    Catching up a bit on the topic...

    > Yes, if the organism has two copies of that gene, it could cover
    >for the fact that most mutations are fatal.<

    This is a popular error. Most mutations have no effect on the
    organism. Of those having small effects, positive and negative are
    about equally common. Fatal mutations get more attention because of
    their large effect.

    >Larry:
    >Whitely, being a rather pure mathematician, was dealing with very
    >general, non-specific cases of search programs. And he was
    >assuming that the population of critters being modeled had no
    >information of their own fitness landscape, and so could only tell
    >whether each mutation eventually helped their survival or not.<

    >One of the computer people in the audience said that he used very
    >special algorithms for each organism he was running, and these
    >usually converged quite rapidly on the target area of the genome
    >space. But then someone else asked him if he was not unfairly
    >putting a lot of privileged information into his algorithm. In
    >that case he is getting out of the calculation the information that
    >he himself put in. The first guy said that's true, but it gets
    >you there fast.<

    My analyses are not optimizing a gene sequence, but rather seek to
    find an optimum network connecting known sequences (i.e., an
    evolutionary tree). The basic process is simple: Construct a
    network. Modify the network in various ways. If a better network is
    found, modify that network and try to improve on it. Stop when no
    better networks are found. Repeating the process several times
    helps. This finds greatly improved networks (though not necessarily
    the best network) much more efficiently than simply trying random
    networks. This seems to directly contradict the No Free Lunch
    principle. It does not involve any knowledge of the goal on the part
    of the program or programmer, and is applicable to any similar data
    set.

         Dr. David Campbell
         Old Seashells
         University of Alabama
         Biodiversity & Systematics
         Dept. Biological Sciences
         Box 870345
         Tuscaloosa, AL 35487 USA
         bivalve@mail.davidson.alumlink.com

    That is Uncle Joe, taken in the masonic regalia of a Grand Exalted
    Periwinkle of the Mystic Order of Whelks-P.G. Wodehouse, Romance at
    Droitgate Spa



    This archive was generated by hypermail 2b29 : Fri Jul 26 2002 - 21:37:29 EDT