The LEM algorithm revolutionizes EC, allowing software-based brain building
I have a friend at George Mason University, VA, USA, Prof. Ryszard Michalski who recently discovered a machine learning based evolutionary algorithm, called LEM (Learnable Evolution Model) http://www.mli.gmu.edu/papers/LEM-MLJ.ps which speeds up evolutionary algorithms, and hence brain building, by a factor of hundreds. I stayed with Ryszard at his home over the summer of 2002 and learned of his LEM. I got excited because it dawned on me that if LEM can make the evolution of individual neural net circuit modules faster by a factor of hundreds, then it would become practical for students to evolve hundreds of them in minutes each and then assemble them into artificial brains. Students could thus build their own aritificial brains - admittedly not with 10,000s of modules or more as would be possible with an evolvable hardware-based brain building machine (similar to the 1st generation brain-building machine "CBM"), but at least with 100s of modules evolved quickly enought to be practical, in software.
The prospect of software-based brain-building became the basic idea of a PhD level course I am now teaching (spring semester of 2003) at USU. To see the powerpoint lecture notes of this course as it unfolds 3 times per week, check out
http://www.cs.usu.edu/~degaris/lecture-notes.html and click on the CS7910 course link. (Use Microsoft Explorer, not Netscape (which formats poorly)).
This is the first such course in the world as far as I know. It will be interesting to see what the students can do with 100s of evolved neural net circuit modules, each with its own function (and hence fitness definition). If this kind of brain-building catches on, then hopefully, there will be enough computer science departments around the world building artificial brains, that soon workshops and later conferences and journals, can be devoted to this new specialty.
For a syllabus of the new software-based brain-building course, CS7910, hit the same website above.