The board game Go is an interesting research object in computer science because currently even the best Go programs are no match for a good human amateur player. Using artificial neural networks is one approach for creating computer Go players. The existing Go software GOjen supports neural network players. While the improvement of their playing abilities through training was already implemented, the project task was to use evolutionary concepts to achieve this goal. Code for evolution of neural networks was already existing, although not for Go software but for SIMMA, a robot simulator. The objective was to adapt the classes responsible for evolution so that they could be used in GOjen.
The Work Document is available as PDF document.