which works particularly well for classification problems with a large number of
classes. Another contribution is the implementation of the system (available on
request).
Future work includes applying the algorithm to demanding incremental clas-
sification problems, for example, web page prediction based on analysis of click
streams or automatic text categorization. Algorithmic improvements that need
to be done include (1) develop balancing mechanisms (in order to give hints for
pivot elements to the applied linear system solver for reduction of numeric er-
rors), (2) add support for decay coefficients for efficient decremental unlearning,
(3) investigate the appropriateness of parallelized incremental proximal SVMs,
(4) strengthen implementation with support for tuning set, kernels as well as
one-against-one classifiers.
References
1. Burbidge, R., Buxton, B.F.: An introduction to support vector machines for data
mining. In Sheppee, M., ed.: Keynote Papers, Young OR12, University of Notting-
ham, Operational Research Society, Operational Research Society (2001) 315
2. Huang, J., Shao, X., Wechsler, H.: Face pose discrimination using support vec-
tor machines (svm). In: Proceedings of 14th Int'l Conf. on Pattern Recognition
(ICPR'98), IEEE (1998) 154156
3. Muller, K.R., Smola, A.J., Ratsch, G., Scholkopf, B., Kohlmorgen, J., Vapnik, V.:
Predicting time series with support vector machines. In: ICANN. (1997) 9991004
4. Fung, G., Mangasarian, O.L.: Incremental support vector machine classification.
In Grossman, R., Mannila, H., Motwani, R., eds.: Proceedings of the Second SIAM
International Conference on Data Mining, SIAM (2002) 247260
5. Fung, G., Mangasarian, O.L.: Multicategory Proximal Support Vector Classifiers.
Submitted to Machine Learning Journal (2001)
6. Schwefel, H.P., Wegener, I., Weinert, K., eds.: 8. Natural Computing. In: Advances
in Computational Intelligence: Theory and Practice. Springer-Verlag (2002)
7. Hettich, S., Bay, S.D.: The UCI KDD archive. http://kdd.ics.uci.edu (1999)
114
Incremental Multicategory PSVM Classifiers