Open Internet by MindsNet

Improving AlphaZero Model Performance on Othello

The author is struggling to improve the performance of their AlphaZero model on a 6x6 Othello board, despite adjusting hyperparameters. The model is not learning to predict values and has a low win rate against a greedy agent. The author seeks to understand the statistical properties of the training data that may be contributing to this issue.

Computing & Technology, Computer Science, Machine Learning