Quiz 4

Milestones

Rubric

All questions on the quizzes in this course are out of 3 points. When assessing these quizzes, score answers as follows:

• 3 is reserved for excellent responses that that fully explain the answer to the question, show a true understanding of the question being asked, and don’t contain any errors. As a general guideline, we expect no more than 25-30% of answers are likely to receive a 3 on any particular question.
• 2 is awarded for a good response that’s getting at the right idea, but might not explain it fully or might have a few minor errors. 2, indeed, is what the course expects most students are likely to earn. As a general guideline, we expect approximately 50-60% of answers are likely to receive a 2 on any particular question.
• 1 is awarded for an inadequate response that misses the mark, is wholly incorrect, or otherwise fails to answer what the question is asking.
• 0 is given to answers that are exceedingly brief, or blank, and as such are effectively non-responses.

The above rubric should be applied for all questions on the quiz.

If you give a score lower than full credit, please include a brief comment in the “Provide comments specific to this submission” section of the form. You can also re-use comments in the “Apply Previously Used Comments” to save time so that you don’t find yourself typing the same thing over and over!

Grading Responsibilities

Answer Key

Question 1

1. Reinforcement learning. The robot’s positive experiences (walking) and negative experience (falling over) allow it to learn what actions to take in the future.
2. Supervised learning. Given labeled data (emails that are spam or not), the AI learns to classify emails into one of two groups.
3. Unsupervised learning. Given just the data about the proteins, the AI learns patterns (structural similarities) that appear in the data.
4. Supervised learning. Given labeled data (photos that are tagged with people), the AI learns to classify photos of people in untagged images.

Question 2

1. L1 Loss = 2 + 1 + 1 + 3 + 1 = 8
2. L2 loss = 2^2 + 1^2 + 1^2 + 3^2 + 1^2 = 16

Question 3

1. 0-1 is most appropriate for classification: we care only about getting the category right or wrong, so subtracting the actual and expected values make less sense. L1 and L2 are most appropriate for regression: we care not just about getting the answer exactly right, but also want to prefer answers that are close to correct over answers that are far from correct.
2. It’s possible that Hypothesis 1 could overfit on the data: resulting in lower loss on the training data, but generalizing less well to the testing data.
3. Regularization is useful because it penalizes more complex hypotheses that might overfit on the data, instead preferring a simpler hypothesis.

Question 4

1. Setting ε = 0 will result in always greedily choosing the current best move estimate, so the AI might never explore parts of the state space that could lead to better moves.
2. Setting ε = 1 will result in always randomly choosing moves, which will result in the AI taking much longer to train, since it will spend time exploring actions we’ve already determined to be poor choices.