Friday, January 18, 2008

Looking back into ANAS

ANAS is John Holland's 1975 monograph, Adaptation in Natural and Artificial Systems. I went back to it recently to review the history of the thinking about linkage learning from the very beginning.

I was struck that Holland's first presentation of a simple genetic algorithm, called a reproductive plan in ANAS, is a steady state GA. In steady state GA, a single individual is replaced at a time in the population. This means that new mixes of genetic material is made available immediately for selection.

I had mis-remembered this. I thought that historically GAs started as generational models in which the entire population was replaced at once, and that steady state GAs were due to Grefenstette.

Similarly, ANAS considers the possibility of evolving algorithms, not just parameter sets. This is the basic distinction between genetic algorithms and genetic programming, which most people attribute to John Koza.

There is a phrase in Pirkei Avot (Ethics of the Fathers) "hufach bo v'hufach bo d'kulla bo." Turn it over, turn it over, because everything is in it. I think it applies just as well to basic texts in the sciences as it does to the Talmud.

No comments: