@PhdThesis{Franc-Phd-TR-2005-22,
  IS = { zkontrolovano 30 Dec 2005 },
  UPDATE  = { 2005-12-14 },
  author =	 {Franc, Vojt{\v e}ch},
  title =	 {Optimization Algorithms for Kernel Methods},
  school = 	 {Center for Machine Perception, 
                  K13133 FEE Czech Technical University},
  address =	 {Prague, Czech Republic},
  year =	 {2005},
  month =	 {November},
  day =          {22},
  type =	 {PhD Thesis},
  number =	 {{CTU--CMP--2005--22}},
  issn =	 {1213-2365},
  pages =	 {119},
  figures =	 {11},
  supervisor =   {Hlav{\' a}{\v c}, V{\' a}clav},
  authorship =	 {100},
  psurl =	 {[Franc-TR-2005-22]},
  project =	 {MSM6840770013, COSPAL IST-004176,
                  GACR 102/03/0440, CONEX GZ 45.535,
                  INTAS 04-77-7347 },
  annote = { This Ph.D. thesis concentrates mainly on the Support
    Vector Machines (SVM) learning of classifiers. Learning of the SVM
    classifier is expressed as a specific convex Quadratic Programming
    (QP) task. The first part of the thesis contributes to the QP
    optimization by proposing new solvers based on the known
    algorithms for the Minimal Norm Problem and the Nearest Point
    Problem. The proposed QP solvers outperformed the state-of-the-art
    methods on a wide variety of problems.  A novel greedy algorithm
    for approximation of the training data embedded in the Reproducing
    Kernel Hilbert Spaces is proposed in the second part of the
    thesis. The method is called Greedy Kernel Principal Component
    Analysis (Greedy KPCA). The Greedy KPCA was successfully applied
    for reduction of complexity of functions learned by the kernel
    methods.  A novel method which allows to transform the learning of
    the multiclass SVM to the singleclass SVM classifier is proposed
    in the third part of the thesis. The transformation is based on a
    simplification of the original problem and employing the Kesler's
    construction. The entire transformation is performed solely by a
    specially designed kernel function.  The proposed methods were
    incorporated to the Statistical Pattern Recognition (STPR) toolbox
    (http://cmp.felk.cvut.cz/~xfrancv/toolbox) written in
    Matlab. A substantial part of the toolbox was designed and
    implemented by the author of the thesis. },
  keywords =	 {kernel methods, support vector machines, 
                  quadratic programming,
                  kernel principal component analysis,
                  multiclass classification },
}