Brief

A user is training an AlphaZero model for Othello on a 6x6 board and encountering issues with performance. Despite models improving against each other, they are not significantly better than benchmark agents, with a win rate below 10% against a greedy agent. The user has analyzed training data, including value loss, prediction entropy, and policy divergence, and is seeking advice on hyperparameter tuning to resolve the model's poor performance. AI