What are the key differences between the three main machine learning paradigms we have seen ( supervised, unsupervised, and reinforcement learning)?

• Supervised learning uses labeled data, unsupervised learning uses unlabeled data, and reinforcement learning involves an agent learning through trial-and-error interactions.

• Supervised learning performs classification or regression, unsupervised learning does clustering, and reinforcement learning maximizes cumulative rewards.

• Both A and B

Which of the following is NOT a Machine Learning approach?

• Supervised Learning

• Unsupervised Learning

• Reinforcement Learning

• Predictive Learning

In the Pong game example, what represents the 'State'?

• The score of the game

• The paddle movement

• The image of the game (pixels)

• The position of the ball only

What technique encourages exploration in Q-learning?

• Random Forest

• Epsilon-Greedy

• Convolutional Networks

• Bellman Equation

In Q-learning, what does the discount factor (γ) represent?

• The learning speed

• The importance of future rewards

• The probability of random actions

• The accuracy of predictions

Which mathematical framework is used in RL to model decision-making processes?

• Convolutional Neural Networks

• Markov Decision Processes (MDP)

• Gradient Descent Algorithms

• Linear Regression

What happens when the agent uses an ε-greedy strategy with ε = 0?

• It explores all possible actions randomly

• It exploits the learned Q-values only

• It fails to update its Q-table

• It stops learning new policies

What is the main difference between Policy-Based and Value-Based RL approaches?

• Policy-based RL estimates Q-values, while Value-based RL searches for policies.

• Policy-based RL focuses on learning the best policy directly, while Value-based RL estimates value functions.

• Policy-based RL is slower than Value-based RL in all environments.

• There is no difference; both rely solely on Q-tables.

When using Deep RL in games like Pong, why is preprocessing the input important?

• To improve the game's graphics

• To reduce computational load and focus on relevant data

• To enhance the randomness of the agent’s actions

• To create a larger state space for better learning

What is the purpose of Epsilon-Greedy Decay?

• To stop exploration completely

• To encourage continuous random actions

• To reduce exploration over time

• To enhance supervised learning models