Mid-term
EECS 545: Machine Learning
Winter, 2018
Name:
UM uniqname:
• Closed books. One sheet of paper of notes is allowed. No computers or calcula-
tors.
– Showing your work makes partial credit possible.
If you write nothing at all, it’s hard to justify any score but zero.
– Feel free to use the backs of the sheets for scratch paper.
– Write clearly. If we can’t read your writing, it will be marked wrong.
• This course operates under the rules of the College of Engineering Honor Code. Your signature
endorses the pledge below. After you finish your exam, please sign below:
I have neither given nor received aid on this examination, nor have I concealed any violations
of the Honor Code.
DO NOT WRITE BELOW THIS LINE
Problem 1 2 3 4 5 6 7 8 Total
Points
Max Points 24 6 12 6 6 6 6 6 72
, 2
Question 1 [24pts] YOU MUST FILL OUT THE SCANTRON SHEET FOR THIS QUESTION
(No need for explanations here unless you feel the question is ambiguous and want to justify your
answer).
1. In general the error on the training set is a better estimate of the generalization error than
the error on the test set.
FALSE
2. The perceptron algorithm finds the maximum margin classifier if the data is linearly separable.
FALSE
3. Locally-weighted linear regression can produce nonlinear fits to the data.
TRUE
4. In nearest neighbor regression, using more neighbors generally leads to more complex func-
tions than using fewer neighbors.
FALSE
5. In polynomial regression with a fixed data set, as one increases the degree of the polynomial
used, the expected mean-squared error on the test set strictly decreases.
FALSE
6. Linear decision boundaries for classification are optimal (minimum misclassification error on
training set) only if the underlying data is truly linearly separable?
FALSE
7. Quadratic discriminant analysis as an approach to classification cannot be applied if the true
class-conditional density for each class is not Gaussian.
FALSE
8. Logistic Regression is a method for doing classification problems.
TRUE
EECS 545: Machine Learning
Winter, 2018
Name:
UM uniqname:
• Closed books. One sheet of paper of notes is allowed. No computers or calcula-
tors.
– Showing your work makes partial credit possible.
If you write nothing at all, it’s hard to justify any score but zero.
– Feel free to use the backs of the sheets for scratch paper.
– Write clearly. If we can’t read your writing, it will be marked wrong.
• This course operates under the rules of the College of Engineering Honor Code. Your signature
endorses the pledge below. After you finish your exam, please sign below:
I have neither given nor received aid on this examination, nor have I concealed any violations
of the Honor Code.
DO NOT WRITE BELOW THIS LINE
Problem 1 2 3 4 5 6 7 8 Total
Points
Max Points 24 6 12 6 6 6 6 6 72
, 2
Question 1 [24pts] YOU MUST FILL OUT THE SCANTRON SHEET FOR THIS QUESTION
(No need for explanations here unless you feel the question is ambiguous and want to justify your
answer).
1. In general the error on the training set is a better estimate of the generalization error than
the error on the test set.
FALSE
2. The perceptron algorithm finds the maximum margin classifier if the data is linearly separable.
FALSE
3. Locally-weighted linear regression can produce nonlinear fits to the data.
TRUE
4. In nearest neighbor regression, using more neighbors generally leads to more complex func-
tions than using fewer neighbors.
FALSE
5. In polynomial regression with a fixed data set, as one increases the degree of the polynomial
used, the expected mean-squared error on the test set strictly decreases.
FALSE
6. Linear decision boundaries for classification are optimal (minimum misclassification error on
training set) only if the underlying data is truly linearly separable?
FALSE
7. Quadratic discriminant analysis as an approach to classification cannot be applied if the true
class-conditional density for each class is not Gaussian.
FALSE
8. Logistic Regression is a method for doing classification problems.
TRUE
