Artificial Intelligence Reinforcement Learning

Question 1

What is the main objective of an agent in reinforcement learning?

Accepted Answer

To maximize cumulative reward over time through interaction with an environment

Answer

To minimize the size of its neural network

Answer

To classify inputs into predefined categories

Answer

To generate synthetic training data

Question 2

What is a 'policy' in reinforcement learning?

Accepted Answer

A mapping from states to actions that defines how the agent behaves

Answer

The set of rules governing the environment's dynamics

Answer

The reward function used to evaluate actions

Answer

The discount factor applied to future rewards

Question 3

What does the 'reward signal' represent in reinforcement learning?

Accepted Answer

Feedback from the environment indicating the immediate value or desirability of the agent's last action

Answer

The learning rate of the neural network

Answer

The total number of states in the environment

Answer

The probability of transitioning between states

Question 4

What is the difference between model-based and model-free reinforcement learning?

Accepted 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

Answer

Model-based RL uses deep networks; model-free RL uses shallow networks

Answer

Model-based RL trains faster; model-free RL is more accurate

Answer

Model-based RL only works in games; model-free RL works in real robots

Question 5

What is an 'episode' in reinforcement learning?

Accepted Answer

A complete sequence of agent-environment interactions from the initial state to a terminal state

Answer

One gradient update step

Answer

A single transition from one state to the next

Answer

The entire training run of an RL agent

Question 6

What is the 'exploration vs. exploitation' dilemma in reinforcement learning?

Accepted Answer

The tension between trying new actions to discover better rewards vs. using known high-reward actions

Answer

Whether to use more compute or less compute

Answer

Whether to train on more data or train longer

Answer

The tradeoff between speed and accuracy in action selection