Thursday, July 28, 2005
A bit of vacation
My IB comrades (hint, hint) will fill in keeping you informed on the latest in the world of genetic algorithms and evolutionary computation. See you after 15 August.
Monday, July 25, 2005
Wednesday, July 20, 2005
MWH Soft extends GA-based scheduler in H20MAP
The new H2OMAP Scheduler extension employs the latest advances in Genetic Algorithm optimization technology with sophisticated Elitist and Global Search Control strategies to automatically determine the optimal pump operation policy for each pump station in the system. By quickly pinpointing the scheduling that best meets target hydraulic and water quality performance requirements, the program facilitates greater productivity with fewer errors, reducing operational costs and maximizing energy savings.A number of years ago, I consulted with MWH Soft, helping them get up to speed on the latest in genetic algorithms by giving a short course for CEO Paul Boulos and a number of MWH Soft engineers.
Monday, July 18, 2005
One little model
Population sizing estimates for GAs based on GR were first provided by Harik, Cantu-Paz, Golberg, and Miller in 1997 and later extended in 1999. The GR problem is simple: start with some initial amount of money, make a series of wagers, and predict how long it will take before you are completely broke or have won $n. There are fixed probabilities for winning or losing each wager.
This can be carried over quite nicely to GAs by considering the pairwise BB-decision making problem - where we are wondering which of two BBs provides the higher fitness. In this case, we consider that the best and second best are battling it out, and that the observed better of the two BBs corresponds to the outcome of the wager. Now in reality, other BBs are also concurrently competing and the probabilities are not fixed, but with some algebraic substitutions and minor approximations, a nice model is derived that is both intuitive and accurate for many problems. See the papers or David Goldberg's book for more information. This has been extended to account for exogenous noise and other tournament sizes. In the original analysis, these factors were not explicitly modeled because the goal was to understand how solution quality and population sizes are related, assuming that the other factors were favorably constant. With a proper population sizing bound, other pertinent relationships can be better understood and modified.
There are a myriad of techniques to derive such little models. These models can explain critical bounds or key trends and behaviors. More complex and accurate models may be needed at times, but little models are often extensible because they capture the underlying concepts of the problem. Anyway, just some musings on an important subject after rereading the GR problem.
Friday, July 15, 2005
IlliGAL Sailing Convention and Seminar a success
Thursday, July 14, 2005
Multimedia art show invokes GAs
'Contagious Brain Blurbs' is a multi-media show about the trajectory of the meme lifecycle: 'encoding and dispersal', 'decoding', 'storage' and 'reinfection.' The works are a series of models illustrating the evolution of memes that are employing genetic algorithms and virtual ecosystems as well as highlighting their mutational process through their change in media or concept from an original form to a possibly evolved or degraded new form.
More information is available at Polvo site.
Wednesday, July 13, 2005
First Annual IlliGAL Sailing Convention & Seminar
Transaction costs and Coases's penguin
An easy way to motivate thinking about transaction costs institutions is the following. If free markets are so great, why don't we wake up every morning and sell our services to the highest bidder? Most of us don't do this. We work for some organization, a company, a university, a government agency, and we agree to work for that same organization, day in and day out, until we change jobs. The answer to the question from transaction cost economics is that there exist costs of using the market (using free markets is not free), transaction costs, that are related to arranging, vetting, and contracting for exchange. These costs are not insubstantial and their existence encourages the formation of the quasi-permanent arrangements (institutions) that we work im.
Coase's Penguin takes transaction cost thinking and applies it to open-source software. Specifically, the article gives a detailed transaction-cost argument for the motivation and emergence of open-source arrangements of creating software. It is well argued, beautifully researched, and nicely written. Give the full article a scan or a read here.
Monday, July 11, 2005
March of the penguins
Saturday, July 09, 2005
Robots are getting personalities.
Friday, July 08, 2005
GECCO-2006 one year away
Strandbeest roams the earth
Thursday, July 07, 2005
Special edition on GAs: Materials & Manufacturing Processes
This collection of 15 papers by eminent academic and industrial researchers from 11 different countries--Australia, Brazil, Finland, India, Netherlands, Poland, Portugal, Russia, Slovenia, United Kingdom and United States--would simply establish that the application of Genetic Algorithms in this area needs to be taken far more seriously than a cult movement destined to appeal only to a select few.Indeed, and besides there is no need to drink cyanide-laced kool-aid to be a part of our uncult. The special edition is a sweet collection for two audiences: material scientists looking for cool GA applications or GA practitioners looking for exemplary applications of GA art to an important field.
Wednesday, July 06, 2005
Southampton researcher receives chair
Dr. Chen's personal page can be examined here.
It was announced last week that four members of the School had been awarded Personal Chairs. In addition to the four new appointments: Professor Sheng Chen, Professor Paul Lewis, Professor Manfred Oper, and Professor Mark Zwolinski, Dr Hugh Davis, Head of the Learning Technologies Group in the School, has been appointed as University Director of Education.
Professor Sheng Chen is a member of the Communications Group. His research interests include adaptive signal processing for communications, modelling and identification of nonlinear systems, learning theory and neural networks, finite-precision digital controller design, evolutionary computation methods and optimization.
Doonesbury v. bloggers
Monday, July 04, 2005
GECCO-2005 proceedings worth a look
The conference and the proceedings are as diverse and interesting as any in recent memory. In talking to Darrell Whitley about the conference, he commented on how it is difficult to get your arms around all the interesting work going on in the field. Ten years ago, people wondered if the field had much steam left in it. The answer is in, and it is a resounding, "yes."
The folks over at IlliGAL have just gotten back from GECCO. Looking through the TOC from the proceedings is making me add another book to the wish list, as well as causing me to bemoan, at least to a small extent, how busy I am. I've had a project on the back burner for some time: A genetic algorithm for fast point-based character generation for the Shadowrun RPG. I know, I know, it is really nerdy, but still, it sounds interesting to me. Unfortunately, I haven't had the time for side projects recently.
I attribute some of this vibrancy to the demewise structure of GECCO. Initiated six years ago at the first GECCO in 1999, we created a structure of separate reviewing for separate areas of the field, and have taken pains to create new demes as new areas or interests arose. GECCO's structure expanded upon a less formal, yet demewise reviewing process initiated in the old GP conference (starting in 1996). This kind of review structure, I believe, encourages differences of opinion that result in exploration of new areas. Of course, doing so merely puts genetic algorithm principles into an organizational context, but earlier GA conferences had a tendency to filter out the unorthodox or unaccepted quite strenuously, even when a vocal minority had embraced and pursued a set of ideas with some vigor.
In IP geek circles, Manfred is legendary; he's the guy who patented the business practice of moving your e-business somewhere with a slack intellectual property regime in order to evade licensing encumbrances. He's the guy who patented using genetic algorithms to patent everything they can permutate from an initial description of a problem domain – not just a better mousetrap, but the set of all possible better mousetraps. Roughly a third of his inventions are legal, a third are illegal, and the remainder are legal but will become illegal as soon as the legislatosaurus wakes up, smells the coffee, and panics.
Does Manfred wear striped shirts, too?