📗 The lecture is in person, but you can join Zoom: 8:50-9:40 or 11:00-11:50. Zoom recordings can be viewed on Canvas -> Zoom -> Cloud Recordings. They will be moved to Kaltura over the weekends.
📗 The in-class (participation) quizzes should be submitted on TopHat (Code:741565), but you can submit your answers through Form at the end of the lectures too.
📗 The Python notebooks used during the lectures can also be found on: GitHub. They will be updated weekly.
📗 Exam:
➭ Exam: PDF, Solution: PDF.
➭ Clarifications:
(1) Question 10: dxy > 0.8 will highlight 2 pixels; dxy > 1 will highlight 0 pixels since none of the pixels are strictly larger than 1; dxy > 0.7 will highlight 3 pixels (0.75, 1, 1).
(2) Question 14: the dual problem should have constraints A' @ y >= c not A' @ y <= c: it's a typo, but it should not affect the answer.
(3) Question 16: probability of transitioning from 0 to 2 is 0, so it is impossible to observe a sequence [0, 0, 2], meaning the probability is 0.
(4) Question 17: lr.predict_proba(x) for multi-class classification returns the probabilities that y belongs to each class, here the probability x belongs to class 0 is 0.3, class 1 is 0.5 and class 2 is 0.2, so lr.predict(x) just returns the class index with the highest probability, i.e. class 1. If you selected the answer 2 and noted in Question 20 that you assumed the classes are 1, 2, 3 not 0, 1, 2, you will get the point back.
(5) Question 19: u1 is principal component 1, u2 is principal component 2, and u3 is principal component 3, and they should not be reordered. This means the reconstruction x is y1 u1 + y2 u2 + y3 u3 = -1 [0, 0, 1] + 0 [1, 0, 0] + 1 [0, 1, 0] = [0, 1, -1].
📗 Exam Coverage:
➭ 20 multiple choice questions (four choices, only one of them is the "most correct").
➭ Week 9 Friday (Nov 3) to Week 15 Wednesday (Dec 13).
📗 Past Exam:
Ten questions are similar to past exams Link or quiz questions (other questions on these past exams are not covered this semester):
📗 New Questions:
Ten questions are not similar to past exam questions or quiz questions, including on the new topics covered this semester:
➭ - Preprocessing: text preprocessing, image preprocessing
Question 1:
➭ - Classification: support vector machines, neural network
Question 2:
Question 3:
➭ - Optimization: linear programming
Question 4:
Question 5:
➭ - Simulation: all lectures
Question 6:
Question 7:
➭ - Enter a different ID to see a different version of the questions: , and click (or hit the "Enter" key).
➭ To check your answer, click :
***** ***** ***** ***** *****
***** ***** ***** ***** *****
➭ There are no more new questions on these topics other than these seven and the ones in the weekly quizzes.
📗 Not on exam:
The following topics are NOT on Exam 3:
➭ Regex (already covered in Exam 2)
➭ HOG Features
➭ Classification methods other than logistic regression, support vector machines, and neural networks
➭ Regression methods other than linear regression
➭ LU decomposition
➭ Statistical inference (t-stats, p-value)
➭ Non-linear optimization methods other than gradient descent
➭ Numerical gradient and hessian
➭ Rand index and adjusted rand index
➭ Non-linear PCA
➭ Column space and row space (not covered this semester)
➭ Tensor operations, including broadcasting (not covered this semester)
➭ Process and thread parallelism (not covered this semester)
1,2,3,4,5,6,7q
📗 [1 points] Suppose dxy = skimage.filters.sobel(img) produces the dxy matrix in the following table. To highlight the edge pixels in the original image in green, image[dxy > t] = [0, 255, 0] is used, and pixels are highlighted. What value of t is used?
0 0 0 0
📗 [1 points] One-vs-one support vector machines are trained and produce the following the confusion matrix. How many training items are used in training the "0 vs 2" support vector machine?
Count
Predict 0
Predict 1
Predict 2
Class 0
Class 1
Class 2
0 0 0 0
📗 [1 points] The 3-fold cross validation accuracy for four different neural networks is summarized below. Which model is the most preferred one based on cross validation accuracy?
Network
Fold 1 accuracy
Fold 2 accuracy
Fold 3 accuracy
A
B
C
D
0 0 0 0
📗 [1 points] What is the optimal solution [x1, x2] to the linear program max c * x1 + x2 subject to x1 + x2 <= 1 and x1 >= 0x2 >= 0 where c = ? 0 0 0 0
📗 [1 points] Suppose the standard form of a linear program max c @ x subject to A @ x <= b and x >= 0 has len(c) = , A.shape = (, ), and len(b) = . What is the number of dual variables len(y)? Note: the dual problem is min b @ y subject to A' @ y >= c and y >= 0 where ' means transpose. 0 0 0 0
📗 [1 points] Suppose all the random vectors generated from a multivariate normal distribution are on the same line, using numpy.random.multivariate_normal([0, 0], [[a, c], [c, b]], 1000) with a = and b = . What is the value of c? 0 0 0 0
📗 [1 points] Consider a Markov chain with the following transition matrix with three states \(\left\{0, 1, 2\right\}\). What is the probability a sequence , , is observed (given it starts with )?