The nature of statistical learning theory
The nature of statistical learning theory
Sparse Greedy Matrix Approximation for Machine Learning
ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning
Probability Estimates for Multi-class Classification by Pairwise Coupling
The Journal of Machine Learning Research
Multiclass reduced-set support vector machines
ICML '06 Proceedings of the 23rd international conference on Machine learning
Solving multiclass learning problems via error-correcting output codes
Journal of Artificial Intelligence Research
Which is the best multiclass SVM method? an empirical study
MCS'05 Proceedings of the 6th international conference on Multiple Classifier Systems
IEEE Transactions on Neural Networks
A comparison of methods for multiclass support vector machines
IEEE Transactions on Neural Networks
The pre-image problem in kernel methods
IEEE Transactions on Neural Networks
Reducing the run-time complexity of support vector data descriptions
IJCNN'09 Proceedings of the 2009 international joint conference on Neural Networks
Fast support vector data descriptions for novelty detection
IEEE Transactions on Neural Networks
Expert Systems with Applications: An International Journal
Computers and Operations Research
| Hi-index | 0.01 |
Multiclass support vector machines (MSVMs) have become a very appealing machine learning approach due to their good results in many classification problems. The resulting machines, though, are usually too large for being usable in many real world applications, especially when fast real-time response is needed. Some approaches aim at decreasing the complexity of the resulting full-size SVM classifiers. The most successful methods follow the ''reduced-set'' procedure, but they have to start from a full SVM solution, and then solve a pre-image problem, prone to fall in local minima. We propose here a compact multiclass SVM (CMSVM) method, that does not need the full SVM solution as a starting point (and hence scales potentially better), and does not need to address the pre-image problem. We evaluate the performance of the proposed scheme by means of real world data sets, and we compare it against other state-of-the-art MSVM techniques.