Support Vector Machines Quiz

Answers and explanations appear as you take the quiz.

1. For linearly separable binary data, what geometric quantity does a hard-margin SVM seek to maximize?
   • The number of deleted training rows
   • The image brightness of every input
   • The separating margin
   • The number of hidden network layers
2. Which training examples contribute as support vectors in a kernel soft-margin SVC decision function?
   • Only correctly classified points nearest the boundary in every case
   • Examples with nonzero dual coefficients, including possible margin violators
   • Every training example with exactly equal weight by definition
   • Only the class label strings without their feature values
3. What computation does a kernel function replace in the SVM formulation?
   • All training labels
   • The requirement for any input examples
   • Every numerical comparison with an alphabetical sort
   • Inner products in an implicit feature representation
4. What does C weight in a standard soft-margin SVM objective?
   • The penalty for margin violations relative to regularization
   • The mandatory number of output colors
   • The file extension of every training example
   • The number of classes in every possible task
5. Which kernel is commonly used to create flexible, curved boundaries?
   • The JPEG kernel
   • The radial basis function (RBF) kernel
   • The TCP kernel
   • The alphabet kernel
6. Which setting makes an SVM a candidate worth evaluating, without guaranteeing its performance?
   • Only inputs with no feature representation
   • A rule that billions of examples are always inexpensive to fit
   • Many input features relative to the number of examples
   • Only audio files with no numerical encoding
7. With a fixed RBF input distance, what does a larger positive gamma do to similarity decay as distance increases?
   • Make it decay more rapidly, producing more localized influence
   • Make it decay more slowly, producing a broader range of influence
   • Make the similarity equal for all nonzero distances
   • Set the penalty for margin violations while leaving similarity unchanged
8. Why can changing feature units affect an RBF SVM unless preprocessing addresses the scales?
   • RBF compares category order while ignoring coordinate distances
   • A constant C automatically cancels every change of feature units
   • A learned intercept always standardizes every input coordinate
   • The kernel uses distances that depend on feature magnitudes