CS 7643 Quiz 5 | New 49 Questions and
Answers | 2026 Update | 100% Correct -
Georgia Tech
This comprehensive study guide contains 49 verified questions and answers for
CS 7643 (Deep Learning) Quiz 5 at Georgia Tech, updated for the 2025/2026
academic year. Topics include Reinforcement Learning, Transformers, Attention
Mechanisms, Semi-Supervised Learning, Meta-Learning, Self-Supervised
Learning, and Generative Models.
SECTION 1: REINFORCEMENT LEARNING (Questions 1-12)
Question 1: What is Reinforcement Learning?
A. A supervised learning method with labeled data
B. Sequential decision making in an environment with evaluative feedback
C. Unsupervised clustering of data points
D. A type of regression analysis
Correct Answer: B
Rationale: RL involves an agent learning to make sequential decisions by
interacting with an environment and receiving evaluative (reward-based)
feedback, without explicit correct action labels .
Question 2: What is Evaluative Feedback in RL?
A. The agent receives a label for the correct action
B. The agent picks an action and receives a reward with no supervision for what
,the correct action would have been
C. The agent receives negative feedback only
D. The agent is told exactly which action to take
Correct Answer: B
Rationale: Unlike supervised learning, evaluative feedback only indicates how
good the action was (reward) without providing the optimal action. This requires
trial-and-error learning .
Question 3: Which of the following is a Signature Challenge in Reinforcement
Learning?
A. Large dataset availability
B. Non-stationarity (data distribution changes when the policy changes)
C. Fixed environment dynamics
D. Immediate rewards for all actions
Correct Answer: B
Rationale: Non-stationarity occurs because as the agent's policy improves, the
distribution of visited states changes. Other challenges include evaluative
feedback, delayed feedback, and fleeting nature of online data .
Question 4: What does MDP stand for in RL, and what does it represent?
A. Markov Decision Process - the framework underlying RL with states, actions,
rewards, and transitions
B. Multiple Decision Protocol - a planning algorithm
C. Maximum Discount Parameter - a hyperparameter
D. Model Distribution Process - for policy evaluation
, Correct Answer: A
Rationale: The Markov Decision Process (MDP) formally defines the RL problem
with components: S (set of states), A (set of actions), R (reward distribution), T
(transition probability), and γ (discount factor) .
Question 5: What is the Markov Property in MDPs?
A. Future states depend on all past states
B. Current state completely characterizes the state of the environment
C. Actions have no effect on state transitions
D. Rewards are deterministic
Correct Answer: B
Rationale: The Markov property states that the future is independent of the past
given the present. The distribution of possible next states depends only on the
current state and action, not previous states or actions .
Question 6: Which equation represents the optimal state-value function
V*(s)?
A. V(s) = max_a [r(s,a) + γ V(s')]
B. V(s) = max_a Σ_{s'} p(s'|s,a) [r(s,a) + γ V(s')]
C. V(s) = Σ_a π(a|s) Q(s,a)
D. V(s) = E[R_t | S_t = s]
Correct Answer: B
Rationale: The Bellman optimality equation for V expresses the maximum
expected return from state s, considering the immediate reward plus discounted
future value averaged over possible next states .*
Answers | 2026 Update | 100% Correct -
Georgia Tech
This comprehensive study guide contains 49 verified questions and answers for
CS 7643 (Deep Learning) Quiz 5 at Georgia Tech, updated for the 2025/2026
academic year. Topics include Reinforcement Learning, Transformers, Attention
Mechanisms, Semi-Supervised Learning, Meta-Learning, Self-Supervised
Learning, and Generative Models.
SECTION 1: REINFORCEMENT LEARNING (Questions 1-12)
Question 1: What is Reinforcement Learning?
A. A supervised learning method with labeled data
B. Sequential decision making in an environment with evaluative feedback
C. Unsupervised clustering of data points
D. A type of regression analysis
Correct Answer: B
Rationale: RL involves an agent learning to make sequential decisions by
interacting with an environment and receiving evaluative (reward-based)
feedback, without explicit correct action labels .
Question 2: What is Evaluative Feedback in RL?
A. The agent receives a label for the correct action
B. The agent picks an action and receives a reward with no supervision for what
,the correct action would have been
C. The agent receives negative feedback only
D. The agent is told exactly which action to take
Correct Answer: B
Rationale: Unlike supervised learning, evaluative feedback only indicates how
good the action was (reward) without providing the optimal action. This requires
trial-and-error learning .
Question 3: Which of the following is a Signature Challenge in Reinforcement
Learning?
A. Large dataset availability
B. Non-stationarity (data distribution changes when the policy changes)
C. Fixed environment dynamics
D. Immediate rewards for all actions
Correct Answer: B
Rationale: Non-stationarity occurs because as the agent's policy improves, the
distribution of visited states changes. Other challenges include evaluative
feedback, delayed feedback, and fleeting nature of online data .
Question 4: What does MDP stand for in RL, and what does it represent?
A. Markov Decision Process - the framework underlying RL with states, actions,
rewards, and transitions
B. Multiple Decision Protocol - a planning algorithm
C. Maximum Discount Parameter - a hyperparameter
D. Model Distribution Process - for policy evaluation
, Correct Answer: A
Rationale: The Markov Decision Process (MDP) formally defines the RL problem
with components: S (set of states), A (set of actions), R (reward distribution), T
(transition probability), and γ (discount factor) .
Question 5: What is the Markov Property in MDPs?
A. Future states depend on all past states
B. Current state completely characterizes the state of the environment
C. Actions have no effect on state transitions
D. Rewards are deterministic
Correct Answer: B
Rationale: The Markov property states that the future is independent of the past
given the present. The distribution of possible next states depends only on the
current state and action, not previous states or actions .
Question 6: Which equation represents the optimal state-value function
V*(s)?
A. V(s) = max_a [r(s,a) + γ V(s')]
B. V(s) = max_a Σ_{s'} p(s'|s,a) [r(s,a) + γ V(s')]
C. V(s) = Σ_a π(a|s) Q(s,a)
D. V(s) = E[R_t | S_t = s]
Correct Answer: B
Rationale: The Bellman optimality equation for V expresses the maximum
expected return from state s, considering the immediate reward plus discounted
future value averaged over possible next states .*
