ML2 Yazeed SVM

What is the intuition behind Support Vector Machines (SVM)?

• To find the average distance between the data points

• To find a hyperplane in a 2-dimensional space

• To minimize the margin between the data points of different classes

• To maximize the margin between the data points of different classes

What is the goal of SVM in terms of margin?

• To minimize the margin between the data points of different classes

• To maximize the margin between the data points of different classes

• To find the average distance between the data points

• To find the smallest minimum distance to the training examples

What are Support Vectors in SVM?

• Data points that are farthest from the hyperplane

• Data points that are closest to the hyperplane

• Data points that do not affect the position of the hyperplane

• Data points that are randomly selected

What is the significance of large margin in SVM?

• It has no impact on the classifier's performance

• It allows for fewer misclassifications and may cause overfitting

• It makes the classifier robust to new examples and contributes to better generalization

• It allows for more misclassifications and may cause underfitting

When are SVMs particularly useful?

• When model training/prediction speed is a priority over accuracy

• In low-dimensional spaces with a small number of features

• When the number of dimensions exceeds the number of samples

• When the data is not linearly separable

What are the practical considerations when using SVM?

• The choice of kernel, the kernel's parameters, and the value of C

• The choice of activation function, the activation function's parameters, and the value of A

• The choice of learning rate, the learning rate's parameters, and the value of L

• The choice of feature space, the feature space's parameters, and the value of D

What is the one-vs-rest approach in Multi-Class SVM?

• Training a single SVM for the majority class

• Training separate SVMs for each class against all the rest

• Training a single SVM for each pair of classes

• Training a single SVM for all classes simultaneously

What is the main problem with the one-vs-rest approach in Multi-Class SVM?

• It is computationally less intensive

• Inputs may be assigned to multiple classes

• It leads to consistent and accurate results

• It is not biased towards the majority class

What is the alternative to the one-vs-rest approach in Multi-Class SVM?

• All of the above

• Single Class SVM

• One-vs-One

• Single Objective Function

What are the pros of SVM?

• Intuitive, can handle non-linear features, computationally less intensive

• Provides convenient probability scores, robust against overfitting in high-dimensional space

• Efficient with a very large number of observations, easy to choose the correct kernel and tuning parameters

• Can handle large feature spaces, adept at handling non-linear feature interactions through kernel trick