Wednesday 26 May 2010

Genetic algorithms

I am ashamed to admit that I was slow to appreciate the possibilities of genetic algorithms. Way back in the 1980's, pre-internet, my colleague Keith and I were visited by a group of researchers from a local compnay who wanted to set up a project with our OR staff at the university. One of them asked if I had thought of using GAs. At that time, I had never heard of them; it was two or three years later that I next encountered them, and I suspect that if we had known even a little about GAs when the project proposal was being put together, we might have been able to contribute some applications in the literature. Since then, some of my work has used GAs and other meta-heuristics.

I was amused to discover the cartoon in xkcd, which imagines what might happen if food recipes were created using GAs.



It reminded me of other examples of cross-over from GAs to everyday life. Ian Stewart imagined evolution of crosses between cats and birds, with an imaginary landscape of various combinations of cats with wings and birds with paws. There have been several research projects for creating abstract art by GAs, and some of these projects have linked the concept to abstract music as well. Somehow, even if we had caught on to the idea of GAs after that casual conversation, I don't think that I would have gone into computer art!

Monday 17 May 2010

Snow, cold and outlying data

One of the recurrent problems of production monitoring is to try and determine if a process is out of control. Observations are made at various times, and from these, one is supposed to determine whether or not there are problems.

An article on the website of KNMI (the Dutch Meteorological Institute) asks whether the winter of 2009-2010 was unusual. It is here. The author asks whether the weather was unusual, by looking at various statistics from different parts of the world. What makes the page so interesting is the way that the statistics are considered with and without a model that incorporates global warming. It is a reminder for O.R. workers doing control modelling to make sure that observations are related to the correct underlying model.