Regularization theory and neural networks architectures
Neural Computation
The nature of statistical learning theory
The nature of statistical learning theory
Machine Learning
Inductive learning algorithms and representations for text categorization
Proceedings of the seventh international conference on Information and knowledge management
An introduction to support Vector Machines: and other kernel-based learning methods
An introduction to support Vector Machines: and other kernel-based learning methods
Maximizing Text-Mining Performance
IEEE Intelligent Systems
Second Order Features for Maximising Text Classification Performance
EMCL '01 Proceedings of the 12th European Conference on Machine Learning
Estimating the Generalization Performance of an SVM Efficiently
ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning
Combining clustering and co-training to enhance text classification using unlabelled data
Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining
Hi-index | 0.00 |
The paper demonstrates that the predictive capabilities of a typical kernel machine on the training set can be a reliable indicator of its performance on the independent test set in the region where scores are larger than 1 in magnitude. We present initial results of a number of experiments on the popular Reuters newswire benchmark and the NIST handwritten digit recognition data set. In particular, we demonstrate that the values of recall and precision estimated from the training and independent test sets are within a few percent of each other for the evaluated benchmarks. Interestingly, this holds for both separable and non-separable data cases, and for training sample sizes an order of magnitude smaller than the dimensionality of the feature space used (e.g. using 驴 2000 samples versus 驴 20000 features for Reuters data). A theoretical explanation of the observed phenomena is also presented.