Machine Learning (TU Wien, Austria)

84 flashcards · Computer Science · by Locke Lamora

1. SVM with gradient descent always finds the optimal hyperplane

   False

2. Gradient descent always finds the global optimum

   False

3. A RNN can be unrolled into an infinite fully connected network

   True

4. Pooling and convolution are operations related to RNNs

   False

5. Learning the structure of a bayesian network is less complex than learning the probabilities

   False

6. SVMs with a linear kernel are especially good for high-dimensional, sparse data

   True

7. Random forest is a homogeneous ensemble method

   True

8. If you use use several weak learners h with boosting to get a classifier H and all h are linear classifiers, then H is also a linear classifier

   False

9. Softmax as activation function is used to scale output to a range of 0..1

   True

10. Kernels can only be used with SVMs

    False

11. Boosting is easy to parallize

    False

12. A good machine learning model has low bias and low variance.

    True

13. Freezing layers means that their weights are only updated during fine-tuning.

    False

14. Leave-p-out cross validation is computationally expensive on large data sets.

    True

15. There exists no Bayesian network where all instatiated nodes are not d-seperated after one node is instatiated.

    False

16. Lasso can not be used for feature selection.

    False

17. F1 used with regression

    False

18. Naive Bayes - probability density function used when only nominal attributes used

    False - only used for numeric

19. MAE is less sensitive to outliers than MSE

    True

20. Overfit is more likely on a smaller test set

    True

21. Paired t-tests used for folds verification in holdout method (train /test split)

    False

22. Random Forests use Bootstrapping

    True

23. Random forests heterogenous ensemble learner?

    False

24. Perceptron with Softmax solves XOR-Problem

    False

25. D-Separation used in neighbourhood searches in a Bayes net

    False

26. Is information gain a method for unsupervised feature selection?

    False

27. Is Bayesian optimization used for construction Bayesian networks?

    False

28. Does the SVM method/algorithm guarantee convergence to globally optimal solution?

    True

29. Lazy learners good for big data

    False

30. Naive Bayes, Laplace correction must be used

    False

31. Paired T-tests used at cross validation

    True

32. when k-nn is used with 1-n encoding, min-max scaling is needed to perform euclid. distance

    False

33. backpropagation is a technique used with perceptrons (single layer)

    False

34. pruning is needed for underfitting

    False

35. k-nn is recommended for large datasets

    False

36. for k-nn categorical features should be normalized

    False

37. Entropy max when all elements from same class?

    False

38. McNemar's Test is used for significance testing?

    True

39. K-d-Tree can be used as search space optimisation for k-NN

    True

40. Random Forests is a boosting ensemble technique

    False

41. Back propagation is a method for training Multi-Layer Perceptrons

    True

42. Ordinal data does not allow distances to be computed between data points

    False

43. In AdaBoost, the weights are uniformly initialised

    True

44. Suppose we have a neural network with ReLU activation function. Let's say, we replace ReLU activations by linear activations. Would this new neural network be able to approximate an XOR function? (Note: The neural network was able to approximate the XOR function with activation function ReLu)

    False

45. The entropy of a data set is based solely on the relative frequencies of the data distribution, not on the absolute number of data points present

    True

46. k-nearest neighbors is based on a supervised learning paradigm

    True

47. Support Vector Machine can by default only solve binary classification problems

    True

48. Naive Bayes gives usually good results for regression data sets

    False

49. Learning the structure of Bayesian networks is usually more complicated than learning the probabilities

    True

50. Chain Rule simplifies calculation of probabilities in Bayesian Networks

    True

51. "Number of attributes of data set" is not a model based features that is used for metalearning

    True

52. Suppose a convolutional neural network is trained on ImageNet dataset. This trained model is then given a completely white image as an input. The output probabilities for this input would be equal for all classes.

    False

53. When learning an SVM with gradient descent, it is guaranteed to find the globally optimal hyper plane.

    False

54. Usually state of the art AutoML systems use grid search to find best hyperparameters

    True

55. Perceptron converges when performed on linearly separable data

    True

56. Perceptron converges when performed on linearly not separable data

    False

57. Gradient boosting minimizes residual of previous classifiers

    True

58. Decision Trees using error rate vs entropy leads to different results

    True

59. Depth of decision tree can be larger than the number of training samples used to create a tree

    False

60. The error of a 1-NN classifier on the training set is 0

    True

61. One-vs-all is an approach to solve multi-class problems for DTs

    False

62. 1-hot encoding is used to transform numerical into categorical attributes

    False

63. The Pearson coefficient has a value range from -1 to 1

    True

64. Off-the-shelf is a transfer learning technique that uses the output of layers from a deep-learning architecture as input for a shallow model

    True

65. SVMs search for a decision boundary with the maximum margin

    True

66. SVMs always find a more optimal decision boundary (hyperplane) than Perceptrons

    False

67. In Bayesian Networks we assume that attributes are statistically independent given the class

    False

68. Majority voting is not used when k-NN is applied for linear regression

    True

69. Chain Rule does not simplify calculation of probabilities for BNs

    False

70. Naive Bayes is a lazy learner

    False

71. Normal Equation (analytical approach) is always more efficient than gradient descent for linear regression

    False

72. knn is based on unsupervised paradigm

    False

73. one vs all is approach used by Naive Bayes

    False

74. Classification is a machine learning task where the target attribute is nominal

    True

75. Decision trees can handle only binary classification problems

    False

76. Macro-averaging for classifier evaluation first calculates accuracy/precision/recall/… per class, before averaging across classes

    True

77. In a data set the entropy is lowest when all classes have the same amount of samples

    False

78. If Naive Bayes is applied on a data set that contains also numeric attributes then a probability density function must always be used

    True

79. Model based features used for meta learning are extracted directly from the data set

    False

80. For the Monte Carlo method in reinforcement learning value estimates and policies are changed only on the completion of an episode

    True

81. Information gain is an unsupervised feature selection method

    False

82. Feature selection is primary useful to improve the effectiveness of machine learning

    True

83. The first model in gradient boosting is a zero rule model

    True

84. PCA is a supervised feature selection method

    False