A male client with a history of heavy alcohol use has been admitted to the hospital for malnutrition and suspected pancreatitis. The client’s diagnostic workup suggests alcoholic ketoacidosis as a component of his current health problems. He is somewhat familiar with the effect that drinking has had on his nutrition and pancreas, but is wholly unfamiliar with the significance of acid–base balance. How best could his care provider explain the concept to him?

The nurse volunteering in the medical tent for a road race on a hot, humid day is asked to see a runner who has collapsed on the road. The nurse notes he has sunken eyes, a temperature of 100°F (37.7°C), and dizziness. These are signs of a fluid volume deficit. Recognizing fluid volume deficit, which of these interventions does the nurse carry out first?

Q16: (4 points)What is the two main objective of a Support Vector Machine?

Q14: (4 points)What is the strategy of AdaBoosting for reweighing the training data points?

Q33: (8 points)After taking the action suggested in the previous question, suppose the discount factor is γ = 0.9, the state transfers from s₂ to s₄ after taking action aₜ, and the reward r is 0.6. Please update the Q-table and write down the updated Q-table. Note: Only one value in the table needs updating, and you might need the Bellman Equation:

Q13: (3 points)In AdaBoosting, there are 8 data instances (i.e., samples). Before updating the weights of the eight samples, the weights of the eight samples are[sample1: 1/8, sample2: 1/8, sample3: 1/8, sample4: 1/8, sample5: 1/8, sample6: 1/8, sample7: 1/8, sample8: 1/8]. After updating the weights of the eight samples, the new weights of the eight samples are[sample1: 0.07, sample2: 0.07, sample3: 0.51, sample4: 0.07, sample5: 0.07, sample6: 0.07, sample7: 0.07, sample8: 0.07]. Which data instances (samples) is/are correctly classified by the decision stump of this iteration?

Q34: (15 points) What are the two key steps of value learning? (4 points) What are the two key steps of policy gradient? (4 points) If we are to build a reinforcement learning with discrete actions, which method we should use? (2 points) What are the weaknesses of value learning (e.g., DQN)? (5 points)

Q28: (6 points)What are the three loss function names of RankSVM, RankBoost, and RankNet discussed in the lectures?

Q25: (10 points) Item-Item Collaborative FilteringBelow is user-movie rating matrix with partial ratings available. Please use the item-item collaborative filtering method to estimate the rating of the user #5 for the movie #1. Hints: first use Pearson correlation as similarity by: subtracting mean rating from each movie, then calculating cosine similarities (Sij) between rows. Later, we predict the rating by taking weighted average using the equation: rix=∑j∈N(i;x)Sij⋅rjx∑Sijr_{ix} = \frac{\sum_{j \in N(i;x)} S_{ij} \cdot r_{jx}}{\sum S_{ij}}​​ where i is the index of an item, x is the index of a user, rjxr_{jx} is the rating of the user x for the item j, N(i;x)N(i;x) is the selected neighbor set of the item i given the user x, and the number of the neighbor set (∣N(i;x)∣|N(i;x)|) is 2.

Q12: (3 points) Answer True or False only: In DQN, the Q value function is instantiated by a neural network.