Artificial Intelligence Reinforcement Learning 1 — Questions and Answers
Question 1: What is the main objective of an agent in reinforcement learning?
- To minimize the size of its neural network
- To maximize cumulative reward over time through interaction with an environment (Correct answer)
- To classify inputs into predefined categories
- To generate synthetic training data
Correct answer: To maximize cumulative reward over time through interaction with an environment
An RL agent learns a policy that maximizes the expected sum of discounted future rewards by taking actions and observing their outcomes in the environment.
Question 2: What is a 'policy' in reinforcement learning?
- The set of rules governing the environment's dynamics
- A mapping from states to actions that defines how the agent behaves (Correct answer)
- The reward function used to evaluate actions
- The discount factor applied to future rewards
Correct answer: A mapping from states to actions that defines how the agent behaves
A policy specifies what action the agent takes in each state, either deterministically (one action per state) or stochastically (a probability distribution over actions).
Question 3: What does the 'reward signal' represent in reinforcement learning?
- The learning rate of the neural network
- Feedback from the environment indicating the immediate value or desirability of the agent's last action (Correct answer)
- The total number of states in the environment
- The probability of transitioning between states
Correct answer: Feedback from the environment indicating the immediate value or desirability of the agent's last action
The reward signal provides scalar feedback after each action, guiding the agent toward behaviors that yield higher cumulative reward over time.
Question 4: What is the difference between model-based and model-free reinforcement learning?
- Model-based RL uses deep networks; model-free RL uses shallow networks
- Model-based RL learns or uses a model of the environment's dynamics to plan; model-free RL learns directly from experience without a model (Correct answer)
- Model-based RL trains faster; model-free RL is more accurate
- Model-based RL only works in games; model-free RL works in real robots
Correct answer: Model-based RL learns or uses a model of the environment's dynamics to plan; model-free RL learns directly from experience without a model
Model-based RL builds a transition and reward model to simulate or plan ahead, while model-free RL learns value functions or policies directly from sampled interactions without an explicit environment model.
Question 5: What is an 'episode' in reinforcement learning?
- One gradient update step
- A complete sequence of agent-environment interactions from the initial state to a terminal state (Correct answer)
- A single transition from one state to the next
- The entire training run of an RL agent
Correct answer: A complete sequence of agent-environment interactions from the initial state to a terminal state
An episode is one full trajectory from start to termination (e.g., a game from start to win/lose), after which the environment resets for the next episode.
Question 6: What is the 'exploration vs. exploitation' dilemma in reinforcement learning?
- Whether to use more compute or less compute
- The tension between trying new actions to discover better rewards vs. using known high-reward actions (Correct answer)
- Whether to train on more data or train longer
- The tradeoff between speed and accuracy in action selection
Correct answer: The tension between trying new actions to discover better rewards vs. using known high-reward actions
An agent must balance exploiting its current best-known actions to get reward versus exploring unknown actions that might yield even higher rewards.
Artificial Intelligence Reinforcement Learning 1
What is the main objective of an agent in reinforcement learning?