Final Submission Version

Version used for final submission. Contains full training pipeline, support for AI vs AI and Human vs AI play, and the training of linear, dense, and convolutional neural networks.

Future work ideas:

• Split put action into placement and ring removal
• Parallelize generation of self play games
• Add additional rotational board symmetries
• Implementation of ResNets
• Further testing on full sized game board
• Further testing of temperature for balancing exploration/exploitation in early turns

Poster Session

Version of project used to conduct experiments and evaluate models for the poster session.