Multivariate Linear Regression on Classifier Outputs: a Capacity Study
We consider the problem of combining the outputs of severed classifiers trained independently to perform a discrimination task, in order to improve the prediction accuracy of individual classifiers. We briefly describe the multivariate linear regression model which has already been implemented successfully for that purpose and we study its capacity, using generalizations of the notion of VC dimension.
KeywordsDiscriminant Function Multivariate Linear Regression Gradient Projection Method Structural Risk Minimization Convex Objective
Unable to display preview. Download preview PDF.
- Freund, Y. and Schapire, R.E. A decision-theoretic generalization of on-line learning and an application to boosting. EuroCOLT’95, 1995, 23–37.Google Scholar
- Sollich, P. and Krogh, A. Learning with ensembles: How over-fitting can be useful. NIPS’8, 1996, 190–196.Google Scholar
- Guermeur, Y., d’Alché-Buc, F. and Gallinari, P. Optimal Linear Regression on Classifier Outputs, ICANN’97, 1997, 481–486.Google Scholar
- Guermeur, Y. An Ensemble Method for Protein Secondary Structure Prediction. Submitted to the Journal of Computational Biology, 1998.Google Scholar
- Guermeur, Y. Combinaison de classifieurs statistiques, application à la prédiction de la structure secondaire des protéines. PhD thesis, Univ. Paris 6, 1997.Google Scholar
- Natarajan, B.K. On learning Sets and Functions. Machine Learning, 1989, 4, 67–97.Google Scholar