| Site Map | Webmaster | ||||
October 7, 1999
NSI in the New York Times
La Jolla, CA: Natural Selection, Inc. was featured today in the New
York Times. In a story titled "Computers That Think Outside the
Box," written by reporter Anne Eisenberg, Chief Scientist Dr. David
Fogel discussed the prospects for using evolutionary computation to
solve difficult problems in engineering and machine learning using only
rudimentary knowledge. Dr. Fogel focused on a collaboration with Kumar
Chellapilla, a Ph.D. student at the University of California at San
Diego, to devise evolutionary algorithms that learn how to play
checkers. "We start with a population of essentially random checker
players and evolve them to become almost as good as human experts
through a process of random variation and selection," Fogel
describes. "What makes this work fundamentally interesting is that
the computer learns how to play simply by playing itself, and without
knowing anything about the features of the game that human experts rely
on." Much like from the movie "War Games," where a
computer taught itself to play tic-tac-toe, Fogel and Chellapilla's
efforts have demonstrated that starting almost from scratch, with little
more than the basic knowledge of how pieces move on the board, an
evolutionary algorithm can extract knowledge as it plays games against
itself and thereby evolve superior strategies. After only 10
generations, the best-evolved checker player was able to defeat both its
creators. Fogel and Chellapilla then took the best-evolved player from
the 100th generation and played it against human opponents on the
internet. The results show that the program was competitive with many
people, and subsequent experiments have earned the program a "Class
A" rating, just one level below "expert," using the
United States Chess Federation rating system.
"It's important to remember here that the evolutionary algorithm had virtually no heuristics for playing good checkers. Whatever it learned, it certainly didn't learn it from me or Kumar: Neither of us are very good at checkers," Fogel says. "This same principle of autonomous evolutionary learning has implications for evolving tactics in other games, and even in competitions between corporations or in military engagements." For more information please see the September 1999 issue of Proceedings of the IEEE, and refer to the paper "Evolution, Games, Neural Networks, and Intelligence," by K. Chellapilla and D. B. Fogel.
Natural Selection, Inc. was founded in 1993 to address complex problems in industry, medicine, and defense. The company possesses unique expertise in computational intelligence techniques, including evolutionary computation, neural networks, and fuzzy logic. The corporation's research efforts support the discovery of new pharmaceuticals, the automated detection and diagnosis of breast cancer from film-screen mammograms, optical character recognition, and a variety of military and industrial projects.
For more information, please contact:
Dr. David B. Fogel,
Executive Vice President & Chief Scientist,
Natural Selection, Inc.,
3333 N. Torrey Pines Ct.,
Suite 200,
La Jolla, CA, 92037
tel: (619) 455-6449
fax: (619) 455-1560
dfogel [at] natural-selection.com
and please visit our web site at www.natural-selection.com.