Go is one notable board game where computer competencestill trails behind that of human experts. In the past,neural-network-based approaches have shown promise. In this paper, theESP variant of the SANE neuro-evolution algorithm was applied to go, andan alternate network architecture featuring subnetworks specialized forcertain board regions was implemented. ESP produced simpler networksthat performed just as well as the more complex ones produced by SANE inother studies. Having region-specialist subnetworks improved the alreadygreat performance marginally. However, both the simple network and thenetwork with specialists failed to scale up to board sizes larger than 7x7.

PDF, PS

Technical Report CSTR01-24, Department of Computer Science, University of Texas at Austin, May 2001. Honors Thesis.

@techreport{perez-bergquist:cstr01-24, title={Applying ESP And Region Specialists To Neuro-Evolution For Go}, author={Andres Santiago Perez-Bergquist}, number={CSTR01-24}, month={May}, institution={Department of Computer Science, University of Texas at Austin}, pages={23}, note={Honors Thesis}, url="http://www.cs.utexas.edu/users/ai-lab/?perez-bergquist:cstr01-24", year={2001} }