Reinforcement Learning(DL) QUIZ (MCQ QUESTIONS AND ANSWERS)

Policy evaluation

Policy optimization

Exploration strategy

Action selection

To store the agent's policy

To store the history of all state-action pairs

To store experiences for training stability and sample efficiency

To store the value function

Exploration strategy

Learning rate annealing

Discount factor adjustment

Policy evaluation

Q-learning

Trust Region Policy Optimization (TRPO)

Proximal Policy Optimization (PPO)

Monte Carlo methods

Policy optimization

Policy evaluation

Value iteration

Temporal Difference (TD) learning

They are computationally efficient

They can handle large state spaces

They are less sensitive to hyperparameters

They are less sample-efficient

Q-learning

Trust Region Policy Optimization (TRPO)

Proximal Policy Optimization (PPO)

Monte Carlo methods

To select actions based on the policy

To estimate the value function

To optimize the policy

To add noise to the policy

Policy gradient

Value function

Advantage function

Q-function

Q-learning

Trust Region Policy Optimization (TRPO)

Proximal Policy Optimization (PPO)

Deep Deterministic Policy Gradient (DDPG)

Model-based methods use value functions, while model-free methods use policy gradients.

Model-based methods rely on a learned model of the environment, while model-free methods do not.

Model-based methods use a deterministic policy, while model-free methods use a stochastic policy.

Model-based methods require less computational resources.

Q-learning

Deep Q-Network (DQN)

Policy gradient methods

Advantage Actor-Critic (A2C)

Value function

Policy gradient

Advantage function

Q-function

Policy optimization

Value iteration

Policy evaluation

Reinforcement learning

Q-learning

Deep Q-Network (DQN)

Policy gradient methods

Monte Carlo methods

To maximize rewards over time

To minimize the state space

To learn from unsupervised data

To generate training examples for a neural network

On-policy methods update the policy while interacting with the environment, while off-policy methods use data from a different policy.

On-policy methods use a deterministic policy, while off-policy methods use a stochastic policy.

On-policy methods rely on value functions, while off-policy methods rely on policy gradients.

On-policy methods are faster in terms of learning convergence.

Q-learning

Deep Q-Network (DQN)

Policy gradient methods

Trust Region Policy Optimization (TRPO)

It leads to overly optimistic estimates of future rewards.

It causes the agent to prioritize immediate rewards over long-term benefits.

It increases the complexity of the policy.

It slows down the learning process.

Balancing the trade-off between learning and taking the best-known action

Determining the agent's policy

Calculating the cumulative reward

Selecting the action with the highest probability

Exploration rate

Discount factor

Learning rate

Policy gradient

Model-free learning

Q-learning

Actor-Critic

Monte Carlo methods

On-policy methods use value functions, while off-policy methods use policy gradients.

On-policy methods learn from the agent's own policy, while off-policy methods can learn from different policies.

On-policy methods require a fixed exploration strategy, while off-policy methods do not.

On-policy methods use a deterministic policy, while off-policy methods use a stochastic policy.

Value function

Policy gradient

Advantage function

Q-function

Policy gradient methods

Q-learning

Actor-Critic

Temporal Difference (TD) learning

Value function

Policy

Q-function

Reward function

Model-based learning

Value iteration

Model-free learning

Q-learning

To determine the agent's policy

To punish the agent for incorrect actions

To specify the agent's exploration strategy

To provide feedback to the agent about its actions

The set of possible actions

The agent's internal memory

The world or system the agent interacts with

The reward function

Supervised learning

Unsupervised learning

Policy optimization

Feature extraction