Module 15: Capstone

Coming soon.

Put it all together. Pick a problem, design a solution, train it, evaluate it, and write it up. This is your portfolio piece.

Project Ideas

1. Train an agent to play Atari Breakout - from pixels to policy.
2. Continuous control with SAC - solve a MuJoCo task and test robustness.
3. Offline RL - apply CQL on a D4RL dataset, compare to behavior cloning.
4. Custom environment - design your own Gymnasium env and train an agent in it.
5. Multi-agent competition - build a competitive multi-agent environment and train competing policies.
6. RLHF mini-project - fine-tune a small language model with a learned reward model.

Deliverables

Coming soon.

Evaluation Criteria

Coming soon.