TU Wien:Machine Learning VU (Musliu)/Exam 2026-01-27

The exam had 75 minutes for answering the question. Three groups. The whole exam consisted of single/multiple choice question.

First section 13 TRUE/FALSE Questions. +2 if correct, -1 if incorrect

Second Section 4 Examples where you have to manually calculate stuff. +3 if correct, -1,5 if incorrect

Third Section Multiple choice with multiple possible correct answers. +2 only if ALL correct answers are crossed. 0 otherwise

Some of the questions were from previous exams. I remember following questions (group B):

(Section 1): **question**: The first model in Gradient Boosting is a zero rule model

(Section 2): One example to calculate the MAE for a regression model 3 + 2*F1 +F2

One example with k-armed bandit

One example to decide, which Feature 1R takes

One example to calculate Naive Baies to predict +/- for an additional row without Laplace correction

(Section 3):

question: Which of the following are well-known CNN architectures?

- a) LeNet -

b) LSTM -

c) ResNet -

d) Reception -

e) TeNet -

f) None of above

question: An output of a convolutional layer is larger when …

- a1) Padding decreases

a2) Padding increases -

b1) Stride decreases -

b2) Stride increases

question: What methods can help with vanishing gradients in neural networks?

5 choices were given, I only remember gradient clipping

question: Which of the following classification methods uses majority voting?

- a) k-NN -

b) Decision Trees -

c) Bayesian Networks -

d) Random Forests -

e) An ensemble of an SVM, Logistic Regression and an MLP with two hidden layers than outputs the prediction of the model with the highest confidence -

f) All of the above -

g) None of the above