SIGEVOlution - Volume 5 Issue 3

EDITORIAL Editorial I t is a great pleasure and a great honor for me to introduce this new issue of SIGEVOlution that hosts an interview with Stewart W. Wilson. To me, Stewart is a mentor and a great friend. To our community, he is the person who, with XCS and XCSF, singlehandedly revolutionized learning classifier system research. In 1994, his paper on XCS in the Evolutionary Computation Journal, brought new life to an almost stalled research field, providing the community with the first classifier system that could tackle a wide variety of difficult problems. In 2001, he introduced XCSF a classifier system for function approximation and opened up a completely new horizon. If you are doing research in learning classifier system today, you are probably using one of his models or something derived from them. The second paper by Sara Silva presents a new recipe for operator equalization to control bloat in Genetic Programming. The paper has been one of the best paper nominees at GECCO-2011 and is reprinted here with the permission of ACM and Sara. An extended version of it also appears in the new Genetic Programming Theory and Practice volume. FOGA-2013 is looking for a home! The deadline to propose a venue to host such an interesting event is September 30 so hurry up! You can find the detailed call for proposal inside the newsletter. Albeit Learning Classifier Systems were proposed by John H. Holland in the 1970s, Stewart is most closely associated with these systems. As the inventor of the zeroth-level classifier system ZCS in 1994 and the accuracy-based classifier system XCS in 1995, he has set a standard in this research realm that is still valid and highly useful today. In fact, the XCS classifier system may be considered one of the most powerful genetics-based machine learning tools available. Originally designed to solve Boolean function problems and maze-like reinforcement learning problems, XCS has now been successfully applied to various other problem domains. In datamining, XCS has shown very robust and highly competitive classification results and has also generated insights into the underlying problem structure. In online generalizing reinforcement learning problems, XCS has successfully solved problems with very huge state spaces (more than 10 27 states). In function approximation problems, XCS has solved high-dimensional regression problems with near optimal final solution distributions. Finally, in the robotics domain, XCS has been shown to autonomously learn forward and …