TU Wien:Machine Learning VU (Musliu)/Exam 2020-09-09

Work time was 90 minutes

1-12 True/False questions

Each correctly answered question was worth 2 points, each wrongly answered question -1. 0 points for no answer.

• Softmax as activation function is used to scale output to a range of 0..1
• Kernels can only be used with SVMs
• Boosting is easy to parallize
• A good machine learning model has low bias and low variance.
• Freezing layers means that their weights are only updated during fine-tuning.
• Leave-p-out cross validation is computationally expensive on large data sets.
• There exists no Bayesian network where all instatiated nodes are not d-seperated after one node is instatiated.
• Lasso can not be used for feature selection.
• Something with "Off-the-shelf" model
• F1 used with regression
• Naive Bayes - probability density function used when only nominal attributes used
• Convolutions and max-pooling layers are important for Recurrent Neural Networks

Bayesian Networks

Describe a local-search based algorithm for creating bayesian networks.

No free lunch theorem

What are the implications of the no free lunch theorem?

Decision

Given was a plot of some 2d data and 4 decision boundaries. Tick which decision boundary was produced by a decision tree.

KNN

Given was a plot of some 2d data and 3 decision boundaries made with KNN for k1, k2 and k3. We should decide which order holds for the ks.

• k1 > k2 > k3
• k2 < k3 < k1
• k1 = k3 = k2
• None of the above

Polynomial Regression

Explain polynomial regression. What are the advantages/disadvantages?

18 Max Pooling

Given was an 7x7 input tensor and a 3x3 layer. Had to tick the right output after max-pooling operation with stride 2.

19 Convolution

Given was an 7x7 input tensor and a 3x3 filter. Had to tick the right output after convolution with stride 2.

20 Naive Bayes

Given was some training and test data. We should calculate the Recall of NB classifier with Laplace Correction.