# Schedule
This is a directed study course in Computer Science. In Spring 2025, the topic is partial observability in multi-agent reinforcement learning. We will implement and compare algorithms used to solve partially observable Markov games: (1) exact methods to solve the belief-state Markov games by (i) discretization of the domain and (ii) iterative elimination of dominated strategies; (2) approximate methods to solve the Markov games where observations are treated as states using deep reinforcement learning techniques.
| Week |
Date |
Topic |
Notes |
| 1 |
- |
Markov Game |
W1 |
| 2 |
- |
Best Response Dynamics |
W2 |
| 3 |
- |
Minimax Q |
W3 |
| 4 |
- |
Partial Observability |
W4 |
| 5 |
- |
Beliefs |
W5 |
| 6 |
- |
Dynamic Programming |
W6 |
| 7 |
- |
Heuristic Search |
W7 |
| 8 |
- |
Project |
W8 |
| 9 |
- |
Project |
W9 |
| 10 |
- |
Project |
W10 |
| 11 |
- |
Project |
W11 |
| 12 |
- |
Project |
W12 |
| 13 |
- |
Project |
W13 |
| 14 |
- |
Project |
W14 |
| 15 |
- |
Project |
W15 |
Notes from previous semesters:
📗 Multi-agent deep learning:
Link 📗 Multi-agent discrete state:
Link 📗 Single-agent deep learning:
Link
Textbooks: (main) Decentralized POMDPs:
Link, Multi-Agent Reinforcement Learning:
Link, Reinforcement Learning:
Link, Game Theory:
Link.