This class focuses on support vector machines (See section 11.4.4 of Luger)
A support vector machine uses a form of competitive learning in which statistical measures determine a minimum set of data points (the support vectors) that maximally separate two classes (classification problem) or the positive and negative instances of a learned concept.
An SVM is a linear classifier and uses supervised learning. In SVM learning the statistical theory of Vapnick and Chervonenkis (VC) is used. The VC dimension is the maximum number of training points that can be divided into two categories by a set of functions. VC theory is used to compute the hyperplane that divides the two classes.
SVM's is a dot product similarity measure to map data from a feature space. Dot product results are linearly combined by weights that are found be quadratic programming.
A kernel function is used to create feature vector mapping onto the support vectors based on statistical regularities in the data. Once trained, new cases are decided by comparison with the support vectors.