How we use Reinforcement Learning (...)
summary
Recreates the 1987 Abbey of the Crime by disassembling the original tape, reimplementing the engine in C++ and exposing it through an OpenAI Gym–style API so agents can play programmatically. Uses random baselines, Markov tables, convolutional neural nets, evolutionary selection and Monte Carlo tree search, stores gameplay with BigQuery + Cloud Functions, runs training on Colab/GPU and Kubernetes/Docker, keeps checkpoints to resume, produces ~15k lines of glue code, runs on a $50/month cloud budget, and admits the full game is not yet solved.
AI-written summary · about
chapters (5)
- 0:00Start
- 22:30CREATE A RL AGENT A RL agent is a program that interacts with an environment, in our case a Open Allgym for Abadla, and learn from observations and rewards
- 24:24REINFORCEMENT LEARNING STRATEGI
- 30:34Game info pipeline
- 39:37Questions?
related talks
32mHow we use Reinforcement Learning to solve the abbey of the crime
28mDeep Reinforcement Learning: theory, intuition, code
31mDesigning game AI with reinforcement learning
37mSave the Princess
02mDesigning a Video Game AI
52mReinforcement Learning for Autonomous Cyber Defense
46mAI Takeover
30mReinforcement Learning - ChatGPT, Playing Games & More