CM50270: Reinforcement learning
[Page last updated: 09 August 2024]
| Academic Year: | 2024/25 |
| Owning Department/School: | Department of Computer Science |
| Credits: | 6 [equivalent to 12 CATS credits] |
| Notional Study Hours: | 120 |
| Level: | Masters UG & PG (FHEQ level 7) |
| Period: |
|
| Assessment Summary: | CW 100% |
| Assessment Detail: |
|
| Supplementary Assessment: |
|
| Requisites: | In taking this module you cannot take CM30359 |
| Learning Outcomes: |
At the end of this unit, students will be able to:
1. describe how reinforcement learning problems differ from supervised learning problems such as regression and classification, 2. formulate suitable real-world problems as reinforcement learning problems by defining a state space, an action space, and a reward function appropriate for the context, 3. critically evaluate a range of basic solution methods to reinforcement learning problems, 4. analyse the difficulties encountered in solving large, complex reinforcement learning problems in practice. |
| Aims: | This unit introduces the reinforcement learning problem and describes basic solution methods. |
| Skills: | Intellectual skills:
* Develop algorithmic thinking for sequential decision making under uncertainty (T, F, A) Transferable skills: * Enhance perspective of decision making (T, F) * Oral presentation of ones work (F,A) |
| Content: | Topics covered normally include: dynamic programming, Monte Carlo methods, temporal-difference algorithms, integration of planning and learning, value function approximation, and policy gradient methods. |
| Course availability: |
CM50270 is Optional on the following courses:Department of Computer Science
|
Notes:
|