sdlee94/Minesweeper-AI-Reinforcement-Learning

Implements Deep Q-Learning with experience replay and epsilon-greedy exploration to train an agent purely through trial-and-error without explicit game rules. Uses a neural network to approximate Q-values across Minesweeper's state-action space, with training across ~500k games demonstrating significant performance improvement. Employs the Bellman equation for value iteration and incorporates hyperparameter tuning (discount factor, epsilon decay) to balance exploration and exploitation during learning.

No commits in the last 6 months.