Optimizing game balance and gameplay with deep reinforcement learning

In this project, the objective is to develop an AI that assumes the roles of both player and game designer. The project is based on bi-level optimization:

1. Player Training: The method learns to play a specific board game with optimal skill within the given rules.

2. Game Optimization: Parallel to learning the game, the system will evolve and modify its rules or setup. The goal is to identify the version of the game that is the most exciting, fairest, and achieves the best game balancing.

PufferLib, a powerful library for Reinforcement Learning, will be used to run millions of simulations in a short time to evaluate the quality of the game and the player's skill.

Qualifications: This project is suitable for students interested in game theory and state-of-the-art AI methods. Students must have completed courses in Artificial Intelligence, Deep Neural Networks, or comparable modules, and possess a strong understanding of programming.