Teach your Pacman to play with ML and Reactive Streams
summary
A Java implementation streams player inputs and agent actions over RSocket into a central knowledge service and trains agents with case-based reinforcement learning (Q-style updates) so Pac-Man tries to maximize survival on an infinite board. The system aggregates concurrent agent streams, stores learned cases instead of full state tables, and runs locally (demo is ML-driven and slower than realtime but functional). Design notes: Monte Carlo, tree search and genetic approaches are evaluated and rejected for scale.
AI-written summary · about
chapters (8)
- 0:00Introduction and setup
- 3:13Why Pac-Man for ML
- 15:59Machine learning approaches
- 29:19Reinforcement learning choice
- 40:22Case-based reasoning
- 50:19Reactive streams overview
- 57:00Demo and implementation
- 1:00:34Summary and resources
related talks
01h 31mReinforcement Learning- Pac-Man
46mAI Takeover
01h 01mReinforcement Learning: Pac-Man
17mTraining My Rival in Java: A Deep Q-Learning AI to Play Azul
56mCan I do AI?
35mLet's play Space Invaders!
21mTraining My Rival in Java: A Deep Q-Learning AI to Play Azul
49mML: Teaching machine to learn from its own mistakes (...)