Empirical Inference Conference Paper 2007

Training and Approximation of a Primal Multiclass Support Vector Machine

We revisit the multiclass support vector machine (SVM) and generalize the formulation to convex loss functions and joint feature maps. Motivated by recent work [Chapelle, 2006] we use logistic loss and softmax to enable gradient based primal optimization. Kernels are incorporated via kernel principal component analysis (KPCA), which naturally leads to approximation methods for large scale problems. We investigate similarities and differences to previous multiclass SVM approaches. Experimental comparisons to previous approaches and to the popular one-vs-rest SVM are presented on several different datasets.

Author(s): Zien, A. and Bona, FD. and Ong, CS.
Book Title: ASMDA 2007
Journal: Proceedings of the 12th International Conference on Applied Stochastic Models and Data Analysis (ASMDA 2007)
Pages: 1-8
Year: 2007
Month: June
Day: 0
Editors: Skiadas, C. H.
Bibtex Type: Conference Paper (inproceedings)
Event Name: 12th International Conference on Applied Stochastic Models and Data Analysis
Event Place: Chania, Greece
Digital: 0
Electronic Archiving: grant_archive
Institution: Max Planck Institute for Biological Cybernetics, Tuebingen, Germany
Language: en
Organization: Max-Planck-Gesellschaft
School: Biologische Kybernetik
Links:

BibTex

@inproceedings{3983,
  title = {Training and Approximation of a Primal Multiclass Support Vector Machine},
  journal = {Proceedings of the 12th International Conference on Applied Stochastic Models and Data Analysis (ASMDA 2007)},
  booktitle = {ASMDA 2007},
  abstract = {We revisit the multiclass support vector machine (SVM) and generalize
  the formulation to convex loss functions and joint feature maps. Motivated by
  recent work [Chapelle, 2006] we use logistic loss and softmax to enable gradient
  based primal optimization. Kernels are incorporated via kernel principal component
  analysis (KPCA), which naturally leads to approximation methods for large scale
  problems. We investigate similarities and differences to previous multiclass SVM
  approaches. Experimental comparisons to previous approaches and to the popular
  one-vs-rest SVM are presented on several different datasets.},
  pages = {1-8},
  editors = {Skiadas, C. H.},
  organization = {Max-Planck-Gesellschaft},
  institution = {Max Planck Institute for Biological Cybernetics, Tuebingen, Germany},
  school = {Biologische Kybernetik},
  month = jun,
  year = {2007},
  slug = {3983},
  author = {Zien, A. and Bona, FD. and Ong, CS.},
  month_numeric = {6}
}