Portfolio Project
Nonogram Solver
Reinforcement Learning (RL)
Context
I wanted to create a machine learning model to automatically solve Nonogram puzzles for me.
Approach
- Built a hybrid CNN + Transformer policy network and trained it on more than 25 million 5×5 puzzles (52,000 episodes × 512-board batches).
- Shaped the reward signal around unique guesses, row/column completions, and full-board solves to speed up exploration.
Impact
- Reached 94% accuracy on unseen 5×5 boards.
