Intelligent Systems II · TEE · papers
Q7(d) Applications of reinforcement learning [5]
Final examination · 17 Apr 2024
Actual historical paper ·100 marks attempted /140 offered · Q1 compulsory + any four of Q2–Q7. Source defects and solution assumptions are explicitly identified.
Discuss the applications of reinforcement learning.
Answer
Reinforcement learning learns a policy for sequential decisions by interacting with an environment and maximizing expected cumulative reward. Unlike supervised learning, each state need not have a supplied correct action label.
| Application | State and action | Reward / reason RL fits |
|---|---|---|
| Robot navigation | State: pose/obstacles; action: movement | Reward for reaching goal, penalties for collision; actions affect future states |
| Game playing | State: board; action: legal move | Win/loss and long-term strategy; a weak immediate move may yield a later win |
| Traffic signal control | State: queue lengths; action: phase/timing | Reduced delay and congestion over time |
| Resource scheduling | State: jobs/load; action: allocation | Throughput and latency rewards under changing demand |
| Recommendation sequences | State: interaction history; action: next item | Long-term engagement rather than only one click |
One worked interpretation: a delivery robot sees state “near junction”, chooses left/right, receives −1 per travel step and +20 on successful delivery. It learns a route balancing path length and successful arrival, using a discount factor for later rewards. Exploration must be controlled when real actions are costly or unsafe; a simulator can provide initial experience. RL is appropriate when actions influence subsequent opportunities, not merely for every static prediction task.