Wednesday, May 04, 2005

Novel Approach To Evolving Neural Networks

Sethuraman Muthuraman, a Malaysian PhD student at Robert Gordon University in Aberdeen Scotland has developed a novel method for evolving neural networks. The system uses an iterative cycle of evolution and structural expansion to the robots body plan.

Muthuraman sees great things arising out of his creation, "I can see that in five to 10 years, intelligent artificial species such as those seen in I.Robot, Terminator and Data from Star Trek will become a part of everyday life." While this may be a bit optimistic seeing how the research is only accomplished simple solutions, this and many other innovations will rapidly propel the global robotic revolution over the next few decades.

The technique is described in more detail Muthuraman's research paper titled, "The Development of Modular Evolutionary Networks for Quadrupedal Locomotion" (PDF).

Via Robots.net