Computer Go -- A Proxy for Key Open Challenges and Opportunities in Computational Intelligence

Missouri University Of Science And Technology, Rolla MO

Investigators

Abstract

Proposal Number: 0725382 Proposal Title: Computer Go -- A Proxy for Key Open Challenges and Opportunities in Computational Intelligence PI Name: Wunsch, Donald C. PI Institution: University of Missouri-Rolla The objective of this research is to illuminate and narrow the differences between computer and human capabilities by making a 9x9 Go player, and creating the groundwork for a 19 x 19 player. The approach is to: Combine Simultaneous Recurrent Networks with Cellular Neural Networks, and train them via Reinforcement Learning, to analyze influence functions; Create Neurofuzzy rules for known patterns; Compare approaches to move filtering; Develop improved tactical analysis; Bootstrap endgame techniques backwards; Develop optimized hardware; Perform outreach. The intellectual merit of the proposed research is: Go is much harder than Chess and its solution offers more to science. Its subtleties mirror core issues in learning. Creative and original concepts are proposed, as outlined in the approach section above. The PI and Co-PI have been collaborators since 1998, and both have significant research track records in many synergistic projects. The broader impacts of the proposed research are: Improved heuristics for cutting through combinatorial complexity. Creating stronger links between learning architectures and improving their training. Contributions to related applications: Economics, Security Applications, Sensor networks, Embedded systems, Biologically-inspired applications, K-12, international, and underrepresented groups outreach. The dissemination of results will be superior, due to the availability of Go rules and data. The project will benefit society, through an improved ability to automate strategic analysis, through better tools, and through an improved workforce.