Announcement

Collapse

Natural Science 301 Guidelines

This is an open forum area for all members for discussions on all issues of science and origins. This area will and does get volatile at times, but we ask that it be kept to a dull roar, and moderators will intervene to keep the peace if necessary. This means obvious trolling and flaming that becomes a problem will be dealt with, and you might find yourself in the doghouse.

As usual, Tweb rules apply. If you haven't read them now would be a good time.

Forum Rules: Here
See more
See less

The latest Bio-Complexity article

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • The latest Bio-Complexity article

    There's an article in the 2017 volume of Bio-Complexity.

    Here's how it starts:
    Source: http://bio-complexity.org/ojs/index.php/main/article/view/BIO-C.2017.1

    Coevolutionary searches are searches where the fitness of a particular solution depends not only on that solution, but also other factors.

    © Copyright Original Source

    That's an incredibly bad start.

    While it may be technically true that coevolutionary searches have changing fitness functions, Ewert and Marks have omitted the basic defining characteristic of a coevolutionary search and the reason why the fitness of a solution cannot be determined just from that solution: that coevolutionary searches are evolutionary algorithms that model two or more populations pitted against each other, and the fitness of an individual organisation is based on how well it competes against one or more members of the other populations, rather than against a fixed standard. Nor can I find any reference to this basic information elsewhere in the paper.

    Ewert and Marks continue:
    Source: ibid

    Such searches have been used in a wide variety of situations. Examples include sorting networks [1], the morphology and performance of competing agents [2], backgammon [3], checkers [4] and chess [5]. While traditional searches require the expertise of penalty function artists to craft a fitness function that guides the algorithm, coevolution is viewed as not requiring this prior expertise.

    © Copyright Original Source

    There's a major problem here - that view applies to all evolutionary search algorithms, not just coevolutionary ones. All that is needed is a means of comparing two potential solutions and determining a winner - and this not only works just as well for multiple competing populations as it does for a single population, but it's trivial to use the same fitness function in both cases. Ewert and Marks don't seem to be aware of this. In fact they don't seem to be aware of the difference between evolutionary and coevolutionary searches at all. That can be determined from this totally false differentiation that seems to assume single-population evolutionary searches need exact fitness functions:
    Source: ibid

    Conventional search algorithms, including evolutionary search, require prior knowledge because of the conservation of information (COI) principles pioneered by Mitchell's analysis of unbiased generalization (1980) ... However, for COI results to apply in the case of conventional search but not in the case of coevolution is odd. Coevolution would appear to be at odds with COI results. What is different about these coevolutionary searches that allows them to not require prior information? The answer is that coevolution queries inferior or subjacent fitnesses that are combined into estimates of full fitness queries.

    © Copyright Original Source

    and much more easily by noting that the reference they cite for checkers doesn't even using a coevolutionary search.*

    While on their checkers cite, Ewert and Marks write this:
    Source: ibid

    Another example is that of checkers [4]. The title of the paper indicates that no human expertise was used in the construction of the strategy. However, the human programmers used a min-max search algorithm as well as adjusting the structure of the neural network to include a piece differential. Both of these are uses of prior knowledge about checkers that human experts have inserted into the search algorithm.

    © Copyright Original Source

    Unfortunately for them, reading the cited paper shows that the prior knowledge is limited to how the game of checkers is played, and doesn't include anything at all about strategy or positional evaluation. Ewert and Marks are complaining that the fitness function for evolving a checkers strategy included having the evolving strategies play checkers against each other, rather than competing at tossing coins, rock-paper-scissors, mahjjong or ten-pin bowling.

    Ewert and Marks have demonstrated once again that the first thing you'll learn if you read their paper is that they know almost nothing about evolutionary algorithms and simulations, have probably never written one, and (as with their article on Steiner trees) may never have even run one.

    Roy

    *"Survival is determined by the quality of play in a series of checkers games played against opponents from the same population."
    Last edited by Roy; 09-07-2017, 10:05 AM.
    Jorge: Functional Complex Information is INFORMATION that is complex and functional.

    MM: First of all, the Bible is a fixed document.
    MM on covid-19: We're talking about an illness with a better than 99.9% rate of survival.

    seer: I believe that so called 'compassion' [for starving Palestinian kids] maybe a cover for anti Semitism, ...

  • #2
    On the bright side, they managed to make that churning engine of ID science, Bio-Complexity, finally break the shutout and publish their first paper in 2017.

    Comment


    • #3
      They're early this year.
      Jorge: Functional Complex Information is INFORMATION that is complex and functional.

      MM: First of all, the Bible is a fixed document.
      MM on covid-19: We're talking about an illness with a better than 99.9% rate of survival.

      seer: I believe that so called 'compassion' [for starving Palestinian kids] maybe a cover for anti Semitism, ...

      Comment


      • #4
        This isn't really an area of my expertise or interest, but just so you don't go completely unchallenged,
        While it may be technically true that coevolutionary searches have changing fitness functions, Ewert and Marks have omitted the basic defining characteristic of a coevolutionary search and the reason why the fitness of a solution cannot be determined just from that solution: that coevolutionary searches are evolutionary algorithms that model two or more populations pitted against each other, and the fitness of an individual organisation is based on how well it competes against one or more members of the other populations, rather than against a fixed standard. Nor can I find any reference to this basic information elsewhere in the paper.
        So basically your complaint is that they should have added one word: “depends not only on that solution, but also other coevolving factors.” Yet by leaving the factors unspecified, they are broadening the range of models in which their conclusions are valid: that is, to the model that is evolving, it doesn’t really matter whether the domain is changing as a direct response to the model’s own changes or to any other related or unrelated factor; what is key is the constantly shifting goalpost. So “the fitness of a particular solution depends not only on that solution, but also other factors” is the preferred summary.


        There's a major problem here - that view applies to all evolutionary search algorithms, not just coevolutionary ones. All that is needed is a means of comparing two potential solutions and determining a winner - and this not only works just as well for multiple competing populations as it does for a single population, but it's trivial to use the same fitness function in both cases.
        So you could program a chess program to teach itself, starting from scratch, to play against a Grand Master (a non-evolving ultimate opponent) by giving it nothing more than the rules of chess and the definition of checkmate? And it's trivial to use the same fitness function as for a program that evolves against an opponent that starts out just as ignorant as itself? What fitness function would you use, without inserting any final information or strategy from the start?
        The whole point of the article is not about what happens if a specific model coevolves with its specific opponent or end goal. The point is merely the question of whether allowing such algorithmic evolution to take place in a gradually evolving environment enables one to bypass the No Free Lunch rule.


        Ewert and Marks don't seem to be aware of this. In fact they don't seem to be aware of the difference between evolutionary and coevolutionary searches at all. That can be determined from this totally false differentiation that seems to assume single-population evolutionary searches need exact fitness functions.
        I think they are using a different definition of coevolution than you are. I would take it that your definition limits it only to two systems that both evolve at the same time and in direct response to each other. They are defining it as a system that must evolve within an evolving domain, in which it starts out with virtually no information, either about the domain or about the eventual goal or ultimate opponent’s strategy. The system then coevolves, not in response a specific rival system, but in tandem with the evolving domain and gradual uncovering of the ultimate goal. A single-population evolutionary search must either be capable of a whole search, where it can look ahead all the way to the goal and figure out how to get there, or be given fitness functions that allow it to grade partial progress, and then it can perhaps tweak and improve the parameters of the fitness function (assuming it at least occasionally achieves its goal) to see if they provide a better gauge of progress toward that goal. The point of the checkers example is that the programmers not only provided the model with the rules of checkers, they provided it with a sample partial progress fitness function and the methods of gauging it, meaning they didn’t start with as blank a slate as they claimed.

        and much more easily by noting that the reference they cite for checkers doesn't even using a coevolutionary search.* "Survival is determined by the quality of play in a series of checkers games played against opponents from the same population."
        it may not use your definition of a coevolutionary search, but it does use the broader definition, a checkers game pitted against a changing environment and improving opponent. Their opponents (not identical to them but chosen from the same population and therefore following the same sort of evolutionary pathway) are evolving, maybe not in direct response to their own play, but in response to the same types of evolutionary pressures.

        Unfortunately for them, reading the cited paper shows that the prior knowledge is limited to how the game of checkers is played, and doesn't include anything at all about strategy or positional evaluation. Ewert and Marks are complaining that the fitness function for evolving a checkers strategy included having the evolving strategies play checkers against each other, rather than competing at tossing coins, rock-paper-scissors, mahjjong or ten-pin bowling.
        No, the prior knowledge also includes the strategy of using min-max searches and a system for analyzing the piece differential that allows for positional evaluation to rank partial progress in situations where the end game is not yet in sight. It gives the model a goal that, evolutionarily speaking, should not be apparent. Using a true blank slate, and a game as complex as checkers with the winning moves not foreseeable at the start, an evolutionary model’s moves should all be utterly random until the first time it wins, and then it can say, Oh, how did I get here? And then evolve its strategy back to the start, and only then realize that some intermediate positions have positional strength.

        Comment


        • #5
          Originally posted by Just Passing Through View Post
          This isn't really an area of my expertise or interest, but just so you don't go completely unchallenged,


          So basically your complaint is that they should have added one word: “depends not only on that solution, but also other coevolving factors.”
          No. They'd still not be mentioning the main characteristic of coevolutionary algorithms - multiple populations. They've completely misdefined a straightforward concept.
          Yet by leaving the factors unspecified, they are broadening the range of models in which their conclusions are valid
          By leaving the factors unspecified they've broadened the term into including a whole host of things it doesn't include under the usual definition - and they've done it without explanation.
          what is key is the constantly shifting goalpost.
          What is key is the multiple populations and inter-population competition that lead to the constantly shifting goalpost. A constantly shifting goalpost without multiple populations would not be coevolution.
          So you could program a chess program to teach itself, starting from scratch, to play against a Grand Master (a non-evolving ultimate opponent) by giving it nothing more than the rules of chess and the definition of checkmate?
          Yes. I think it's been done.
          And it's trivial to use the same fitness function as for a program that evolves against an opponent that starts out just as ignorant as itself?
          It's trivial to use the same fitness function for an evolutionary algorithm with one population vs one with multiple populations.
          What fitness function would you use, without inserting any final information or strategy from the start?
          Play one member of a population against another and keep the one that wins. Same as was done in the checkers paper.
          I think they are using a different definition of coevolution than you are.
          I think they're using a definition of coevolutionary that is not only different from the standard one but misleadingly so to the point that it detracts from their paper. Just like using a definition of "city" as "somewhere you can buy bread and candles" would be.
          I would take it that your definition limits it only to two systems that both evolve at the same time and in direct response to each other.
          I'm not sure you know how evolutionary algorithms work either. Not least because this:
          A single-population evolutionary search must either be capable of a whole search, where it can look ahead all the way to the goal and figure out how to get there, or be given fitness functions that allow it to grade partial progress, and then it can perhaps tweak and improve the parameters of the fitness function (assuming it at least occasionally achieves its goal) to see if they provide a better gauge of progress toward that goal.
          Is false.
          Jorge: Functional Complex Information is INFORMATION that is complex and functional.

          MM: First of all, the Bible is a fixed document.
          MM on covid-19: We're talking about an illness with a better than 99.9% rate of survival.

          seer: I believe that so called 'compassion' [for starving Palestinian kids] maybe a cover for anti Semitism, ...

          Comment


          • #6
            I'm shocked that Bio-Complexity still exists. I think we're in a relatively calm period where creationism isn't too much of a force. This may change in the coming years.

            It's a bit late now, but I'm meeting with one of the leading specialists on creationism as a political/historical phenomenon tomorrow. If you want me to ask a particular question, PM me.

            Comment


            • #7
              Some additional points:
              Originally posted by Just Passing Through View Post
              No, the prior knowledge also includes the strategy of using min-max searches and a system for analyzing the piece differential that allows for positional evaluation to rank partial progress in situations where the end game is not yet in sight. It gives the model a goal that, evolutionarily speaking, should not be apparent.
              This isn't true.

              The min-max search isn't a strategy employed by the evolving neural network, it's part of the game simulator. The neural networks are not given this knowledge.

              For the piece differential, while the difference in the number of pieces each player has is an input into the neural network, it is just a number with no obvious significance, and in particular it is not* fed into the neural networks as something to be minimised. The goal is not apparent.

              You appear to be relying on Ewert/Marks (mis)characterisation of the checkers work whereas the actual paper makes it clear that the neural networks are not given the piece differential as a goal:
              Source: http://ieeexplore.ieee.org/document/942536/

              Suppose you are asked to play a game on an eight-by-eight board of squares with alternating colors. There are 12 pieces on each side arranged in a specific manner to begin play. You are told the rules of how the pieces move (i.e., diagonally, forced jumps, kings) and that the piece differential is available as a feature. You are not, however, told whether or not this differential is favorable or unfavorable ... or if it is even valuable information. Most importantly, you are not told the object of the game. You simply make moves and at some point an external observer declares the game over. They do not, however, provide feedback on whether or not you won, lost, or drew. The only data you receive comes after a minimum of five such games and is offered in the form of an overall point score. Thus, you cannot know with certainly which games contributed to the overall result or to what degree.

              © Copyright Original Source



              Using a true blank slate, and a game as complex as checkers with the winning moves not foreseeable at the start, an evolutionary model’s moves should all be utterly random until the first time it wins, and then it can say, Oh, how did I get here?
              As can be seen from the above, that's effectively what was done - except you've overstated the amount of feedback provided. The neural networks didn't even get feedback on when they'd won.

              *it is a feature of the checkers program they used to test the effectiveness of the evolved neural networks.
              Last edited by Roy; 09-08-2017, 04:24 AM.
              Jorge: Functional Complex Information is INFORMATION that is complex and functional.

              MM: First of all, the Bible is a fixed document.
              MM on covid-19: We're talking about an illness with a better than 99.9% rate of survival.

              seer: I believe that so called 'compassion' [for starving Palestinian kids] maybe a cover for anti Semitism, ...

              Comment


              • #8
                Originally posted by HMS_Beagle View Post
                On the bright side, they managed to make that churning engine of ID science, Bio-Complexity, finally break the shutout and publish their first paper in 2017.
                They may have published, but it is not science.
                Glendower: I can call spirits from the vasty deep.
                Hotspur: Why, so can I, or so can any man;
                But will they come when you do call for them? Shakespeare’s Henry IV, Part 1, Act III:

                go with the flow the river knows . . .

                Frank

                I do not know, therefore everything is in pencil.

                Comment


                • #9
                  Originally posted by psstein View Post
                  I'm shocked that Bio-Complexity still exists. I think we're in a relatively calm period where creationism isn't too much of a force. This may change in the coming years.

                  It's a bit late now, but I'm meeting with one of the leading specialists on creationism as a political/historical phenomenon tomorrow. If you want me to ask a particular question, PM me.
                  This needs explaining, because bio-complexity is simply the nature of the evolving life.

                  There is not a relative calm, like the history of evolution over the millennia. At present the evolving bio-complexity of life is extremely active in the Rain Forests of the Tropics that are not decimated by humans.
                  Glendower: I can call spirits from the vasty deep.
                  Hotspur: Why, so can I, or so can any man;
                  But will they come when you do call for them? Shakespeare’s Henry IV, Part 1, Act III:

                  go with the flow the river knows . . .

                  Frank

                  I do not know, therefore everything is in pencil.

                  Comment


                  • #10
                    Originally posted by shunyadragon View Post
                    This needs explaining, because bio-complexity is simply the nature of the evolving life.

                    There is not a relative calm, like the history of evolution over the millennia. At present the evolving bio-complexity of life is extremely active in the Rain Forests of the Tropics that are not decimated by humans.
                    Er, he's talking about Bio-Complexity, the phony and incestuous ID "science" journal the Discovery Institute puts out. The one that averages 1 1/2 articles per years.

                    Comment


                    • #11
                      Originally posted by HMS_Beagle View Post
                      Er, he's talking about Bio-Complexity, the phony and incestuous ID "science" journal the Discovery Institute puts out. The one that averages 1 1/2 articles per years.
                      This even makes it worse!
                      Glendower: I can call spirits from the vasty deep.
                      Hotspur: Why, so can I, or so can any man;
                      But will they come when you do call for them? Shakespeare’s Henry IV, Part 1, Act III:

                      go with the flow the river knows . . .

                      Frank

                      I do not know, therefore everything is in pencil.

                      Comment


                      • #12
                        I was frankly surprised to see it publish anything. I thought it had dwindled away after the creationists suffered a legal defeat on whether intelligent design was simple creation science rebranded.

                        Comment


                        • #13
                          Originally posted by Leonhard View Post
                          I was frankly surprised to see it publish anything. I thought it had dwindled away after the creationists suffered a legal defeat on whether intelligent design was simple creation science rebranded.
                          It wasn't founded until 2010, five years after that legal defeat. AFAICT it exists only so that IDers can pretend they have published their material in a peer-reviewed form. Though with articles like this it's not much of a pretence.
                          Last edited by Roy; 09-11-2017, 09:19 AM.
                          Jorge: Functional Complex Information is INFORMATION that is complex and functional.

                          MM: First of all, the Bible is a fixed document.
                          MM on covid-19: We're talking about an illness with a better than 99.9% rate of survival.

                          seer: I believe that so called 'compassion' [for starving Palestinian kids] maybe a cover for anti Semitism, ...

                          Comment


                          • #14
                            Originally posted by Roy View Post
                            It wasn't founded until 2010, five years after that legal defeat. AFAICT it exists only so that IDers can pretend they have published their material in a peer-reviewed form.
                            IIRC there are only a small handful involved who essentially "peer-review" each other.

                            I'm always still in trouble again

                            "You're by far the worst poster on TWeb" and "TWeb's biggest liar" --starlight (the guy who says Stalin was a right-winger)
                            "Overall I would rate the withdrawal from Afghanistan as by far the best thing Biden's done" --Starlight
                            "Of course, human life begins at fertilization that’s not the argument." --Tassman

                            Comment


                            • #15
                              Originally posted by rogue06 View Post
                              IIRC there are only a small handful involved who essentially "peer-review" each other.
                              After 8 years of publishing they still haven't had as many authors (25) as there are members of their editorial team (31).
                              Jorge: Functional Complex Information is INFORMATION that is complex and functional.

                              MM: First of all, the Bible is a fixed document.
                              MM on covid-19: We're talking about an illness with a better than 99.9% rate of survival.

                              seer: I believe that so called 'compassion' [for starving Palestinian kids] maybe a cover for anti Semitism, ...

                              Comment

                              Related Threads

                              Collapse

                              Topics Statistics Last Post
                              Started by Hypatia_Alexandria, 03-18-2024, 12:15 PM
                              48 responses
                              135 views
                              0 likes
                              Last Post Sparko
                              by Sparko
                               
                              Started by Sparko, 03-07-2024, 08:52 AM
                              16 responses
                              74 views
                              0 likes
                              Last Post shunyadragon  
                              Started by rogue06, 02-28-2024, 11:06 AM
                              6 responses
                              47 views
                              0 likes
                              Last Post shunyadragon  
                              Working...
                              X