Machine Learning Basics (v) Support vector machines

Machine learning must be familiar to support vector machines (support vector MACHINE-SVM), because before the depth of learning, SVM has been hogging the machine to learn Big Brother's seat. His theory is very beautiful, various varieties of improved version also many, such as LATENT-SVM, STRUCTURAL-SVM and so on. This section first looks at the theory of SVM, in (Fig.) A graph shows two types of data sets, graph b,c,d all provide a linear classifier to classify data? But which effect is better?

(Figure I)

Maybe three classifiers are just as good for this dataset, but in fact, this is just a training set, the actual test sample distribution may be more scattered, all kinds of possible, in order to deal with this situation, we have to do is to make the linear classifier as far as possible from the two datasets are as much as possible, This will reduce the risk of the actual test sample crossing the classifier and improve the detection accuracy. The idea of maximizing the spacing between the dataset and the classifier (margin) is the core idea of the support vector machine, while the nearest sample from the classifier becomes the support vector. Now that we know our goal is to find the maximum margin, how do we look for a support vector? How to achieve it? The following (Figure II) shows how to accomplish these tasks.

(Figure II)

The assumption (figure II) in the line represents a super plane, in order to see the aspect of the display into a one-dimensional line, features are the hyper-dimensional dimension plus one dimension, the figure can also be seen, the characteristics are two-dimensional, and classifier is one-dimensional. If the feature is three-dimensional, the classifier is a plane. The analytic expression of the super surface is assumed to be

(Figure III)