Reinforcement Learning CT-AI Practice Questions (Page 2)

Questions 6–10

1. 6application · easy

   An RL agent plays chess. It has a policy that tells it which move to make in each board position, and a value function that estimates the expected return from each position. Which statement best describes the difference between the policy and the value function?

   • AThe policy is a number; the value function is a set of rules.
   • BThe policy estimates future rewards; the value function selects the next move.
   • CThe policy maps states to actions; the value function estimates the expected cumulative reward from a state.
   • DThe policy is used only during training; the value function is used only during deployment.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A QA team wants to use RL to automatically explore a mobile app to find crashes. The agent can tap, swipe, or enter text. The reward is +1 for reaching a new screen and +10 for triggering a crash. Which of the following best describes this application?

   • AIt is unsupervised learning because the agent explores without labels.
   • BIt is supervised learning because the app provides feedback.
   • CIt is not RL because the agent does not have a clear goal.
   • DIt is a valid RL application because the agent learns to maximize rewards from exploration actions.
   Reveal Solution Discussion Select an answer first
4. 8application · easy

   In an RL system for a self-driving car, the agent's state includes the car's speed, position, and nearby obstacles. The agent can accelerate, brake, or steer. What is the role of the reward function in this system?

   • AIt defines the goal by providing feedback on how good an action was in a given state.
   • BIt determines the next state of the environment.
   • CIt stores past experiences for the agent to replay.
   • DIt selects the action to take in each state.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   An RL agent is trained to play a video game. Early in training, it takes many random actions, but later it mostly follows the policy it has learned. This change is an example of:

   • ADecreasing exploitation over time to avoid overfitting.
   • BIncreasing exploitation over time while still allowing some exploration.
   • CSwitching from model-based to model-free learning.
   • DEliminating exploration entirely after training.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A team is deciding between supervised learning and reinforcement learning for a system that controls a robot arm. They have a dataset of successful and unsuccessful arm movements. Which approach should they choose, and why?

   • AEither approach, because both can use labeled data to learn a policy.
   • BUnsupervised learning, because it can cluster movements into successful and unsuccessful groups.
   • CReinforcement learning, because it can learn from trial and error without labeled data.
   • DSupervised learning, because they have labeled examples of good and bad movements.
   Reveal Solution Discussion Select an answer first