News
-
BEKO FRIDGE-FREEZER good working order
4.2/2.5...
-
BIRD CAGE, 24ins tall, med,
£5 0...
-
Soul Searcher
High-concept, effects-laden drama in which a lowly, young street-sweeper is chosen as the successor to the dying Grim Reaper. This fantastical story was a remarkable four years in the making and made for no more than the price of an average family car. Neil Oseman Ray Bullock Jr. Katrina Cook Lara Greenway
-
Gallery: SEGA Rally (Xbox 360)
-
Evolutionary computation
There are lots of ways of getting a computer to mimicevolution, but the one I like best, and has the cleanest conceptual lines, is the"genetic algorithm"invented byJohn Holland. Take a population of strings, and measure their"fitness"against some criterion. Each string has a random number of descendants, i.e. gets copied a random number of times; the expected number of descendants of a string is proportional to its fitness. These descendants can be mutants, or swap part of their length with another descendant string. (Oral tradition says that this"cross-over"is usually more useful than mutation; when and how much, nobody really knows.) Repeat as needed. For this basic version of the genetic algorithm, Holland proved what is called theschema theorem,showing that, if certain patterns of strings are fitter than others, then, in the long run, the GA can't help but find them, and the average number of strings in the population matching those patterns will increase exponentially. (This result is a relative, and quite consciously so, of Fisher's"Fundamental Theorem of Natural Selection.") In retrospect, it's an almost obvious idea, and utter idiots can write a workable GA in Fortran. (I have.)In the normal computer-science manner, this has of course been tweaked and elaborated in hundreds, and by now probably literally thousands, of different ways. John Koza's technique of"genetic programming"replaces the simple strings with a clever technique for only encoding syntactically correct (though possibly useless) programs. (It only works in languages with a very regular syntax, like Lisp or, may the gods defend us, Forth.) There are even variants where the fitness is determined (more or less) endogenously, i.e. which use something more like natural than artificial selection. (Obviously, though, artificial selection is the way to go if you want to solve a particular problem.) It's neat, it's significant, and if I here it described once more as"teaching computers how to have sex"or"the merging of life and the machine,"I believe I'll scream. Of course itisused inartificial lifeandmachine learning, but if Holland had just picked some dull, boring name, like"stochastic combinatory optimization", nobody would babble about it at all.Recommended:physics/0102009.santafe.edu/~mm/">Melanie Mitchell,Introduction to Genetic Algorithms[I read this and came to the conclusion it's the best intro. book on GAsbeforeI started working for Melanie, honest.]To read:J. Berard and A. Bienvenue,"Sharp Asymptotic Results for Simplified Mutation-Selection Algorithms",The annals of Applied Probability13(2003): 1534--1568Volkhard Buchholtz and Thorsten Poeschel,"Adaptive Evolutionary Optimization of Team Work,"cond-mat/0203412Kenneth A. De Jong,Evolutionary Computation: A Unified Approach[Blurb]Fogel,Evolutionary ComputationGoldberg,Genetic algorithmsHughes Juillé,Methods for Statistical Inference: Extending the Evolutionary Computation Paradigm[Ph.D. dissertation, Brandeis University, 1999;online]Hillol Kargupta,"Information Transmission in Genetic Algorithm and Shannon's Second Theorem"Koza,Genetic ProgrammingLaura Landweber and Erik Winfree (ed.),Evolution as Computation[The dual subject, as were]Heinz Mühlenbein [Website for papers]HM,"The Equation for the Response to Selection and Its Use for Prediction,"Evolutionary Computation5(1998): 303--346HM and Thilo Mahnig,"The Exact Equation for the Response to Selection for Multiple Alleles and Its Application"HM and Alberto Ochoa Rodriguez,"Schemata, Distributions and Graphical Models in Evolutionary Optimization"Juan Pablo Neirotti and Nestor Caticha,"Dynamics of the evolution of learning algorithms by selection",Physical Review E67(2003): 041912Wilhelm Stannat,"On the Convergence of Genetic Algorithms --- A Variational Approach",Probability Theory and Related Fields129(2004): 113--132Marc Toussaint,"Self-adaptive exploration in evolutionary search,"physics/0102009.html">John Holland,Adaptation in Natural and Artificial Systems[One of the best scientific books I've ever read]