Material covered
 Classification based on classconditional density
 Bayesoptimal classification, Bayes risk, discriminants
 Multivariate Gaussian (normal) distribution review, covariance
 Discriminants between Gaussian classconditional densities
 Linear discriminant in the special case of equal covariance for all classes
 Comments on the special case of spherical Gaussian densities
 Discriminants are quadratic surfaces in case of diverse covariances
 Loss functions: 0/1 ("true"), perceptron, hinge, square
 Adding loss functions over instances in the model space
 The perceptron algorithm (proof of convergence deferred)
 Local regression using kernels
 Limiting case of impulse kernels supports largemargin principle

Last modified: Saturday, 25 August 2007, 10:14 AM