| Attempts to solve the game of Go using techniques that have worked well for solving similar games have been met with limited success. One difficulty Go presents is that the size of the search space makes it impractical to store a large number of a players board configurations that are known to be good/bad, or to find if such a configuration occurs in a given board region. By treating the board as a simple gray-scale image, we can apply image recognition techniques which have proven successful in fuzzy matching to determine within a certain degree of certainty not only that a given configuration is actually present, but if a given shape is reachable from the current position.;The Elastic Graph Matching algorithm was implemented, and a suitable distance function created and applied. A testing methodology was then developed and applied to several thousand test cases using various permutations of the parameters to explore the suitability of the algorithm to the problem.;The approach shows promise. Future work might involve finding better values of the parameters, as well as developing additional parameters for improved accuracy. |
