(7th-June-2020)
• A support vector machine (SVM) is used for classification. It uses functions of the original inputs as the inputs of the linear function. These functions are called kernel functions. Many possible kernel functions exist. An example kernel function is the product of original features; adding the products of features is enough to enable the representation of the exclusive-or function. Increasing the dimensionality can, however, cause overfitting. An SVM constructs a decision surface, which is a hyperplane that divides the positive and negative examples in this higher-dimensional space. Define the margin to be the minimum distance from the decision surface to any of the examples. An SVM finds the decision surface with maximum margin. The examples that are closest to the decision surface are those that support (or hold up) the decision surface. In particular, these examples, if removed, would change the decision surface. Overfitting is avoided because these support vectors define a surface that can be defined in fewer parameters than there are examples.
Comments