Quiz16

To use SVMs in non-separable, nonlinear case

To use SVMs in non-separable, nonlinear case

Answer

We map data to \(\mathbb{R}^h\) where h is typically much larger than original dimensionality

Which of the following statements is TRUE about kernel SVM

Which of the following statements is TRUE about kernel SVM

Answer

  • ☐ Knowing \(\phi(x_i) \forall i \in \{1\dots n\}\) is necessary to predict
  • ☐ \(\mathbf{K}(x_i, x_j)\) must be a positive-definite matrix
  • ☐ Knowing \(\mathbf{K}(x_i,x_j), \forall i,j \in \{1, \dots, n\}\) is sufficient to predict,
  • ☑ Knowing \(\mathbf{K}(x_i,x_j), \alpha_i, \alpha_j \forall i,j \in \{1, \dots, n\}\) is enough to predict,

Which of the following is FALSE about kernels

Which of the following is FALSE about kernels

Answer

  • ☐ Kernel matrices are of size \(n \times n\)
  • ☐ Kernels should be symmetric
  • ☐ Kernels should be positive semidefinite matrices
  • ☑ Kernels should not be a dot product of non-linear functions.

Which of the following are FALSE:

Which of the following are FALSE:

Answer

  • ☑ Association rules mining is suited for non-overlapping rules
  • ☐ When rules overlap, we need to determine the ranking
  • ☐ Association rules mining is suited for discrete attributes

Which of the following is FALSE about association rules

Which of the following is FALSE about association rules

Answer

  • ☐ Rules are ranked on the basis of confidence
  • ☐ Class labels are always on the RHS of the rules (a => labels)
  • ☐ Itemsets without labels are irrelevant
  • ☑ Only maximal itemsets are used for learning rules