Mostefa Golea, Peter L. Bartlett, Wee Sun Lee, and Llew Mason. Generalisation in decision trees and DNF : Does size matter? Tobepresented at Neural Information Processing Systems 1997. Home/Search Document Not in Database Summary Related Articles Check |
This paper is cited in the following contexts:
Generalization in Threshold Networks, Combined Decision.. - Mason, Bartlett, Golea Self-citation (Golea Bartlett Mason) (Correct)
....for a particular convex combination of hypotheses the complexity term necessarily depends on the complexity of the most complex hypothesis, even if this hypothesis has a low voting weight and hence little effect on the combined hypothesis. Golea et al. present a generalization of this result in [5] where the bound on generalization error depends on the margin and on a complexity term which involves the average VC dimension of the class of base hypotheses (where the average is in term of the weights assigned to the base hypotheses in the convex combination) In Section 3 of this paper we ....
....is in terms of the weights assigned to the hidden units and the complexity of an individual hidden unit is in terms of the proportion of training examples which are classified near threshold) This is accomplished by considering the network as a thresholded convex combination of hidden units. In [5] Golea et al. bound the generalization error of a binary decision tree in terms of the margin and a complexity term by considering a binary decision tree as a thresholded convex combination of leaf functions. Their complexity term depends on the VC dimension of the class of node decision ....
[Article contains additional citation context not shown here]
Mostefa Golea, Peter L. Bartlett, Wee Sun Lee, and Llew Mason. Generalisation in decision trees and DNF : Does size matter? Tobepresented at Neural Information Processing Systems 1997.
Generalization in Threshold Networks, Combined Decision.. - Mason, Bartlett, Golea Self-citation (Golea Bartlett Mason) (Correct)
....for a particular convex combination of hypotheses the complexity term necessarily depends on the complexity of the most complex hypothesis, even if this hypothesis has a low voting weight and hence little effect on the combined hypothesis. Golea et al. present a generalization of this result in [5] where the bound on generalization error depends on the margin and on a complexity term which involves the average VC dimension of the class of base hypotheses (where the average is in term of the weights assigned to the base hypotheses in the convex combination) In Section 3 of this paper we ....
....is in terms of the weights assigned to the hidden units and the complexity of an individual hidden unit is in terms of the proportion of training examples which are classified near threshold) This is accomplished by considering the network as a thresholded convex combination of hidden units. In [5] Golea et al. bound the generalization error of a binary decision tree in terms of the margin and a complexity term by considering a binary decision tree as a thresholded convex combination of leaf functions. Their complexity term depends on the VC dimension of the class of node decision ....
[Article contains additional citation context not shown here]
Mostefa Golea, Peter L. Bartlett, Wee Sun Lee, and Llew Mason. Generalisation in decision trees and DNF : Does size matter? To be presented at Neural Information Processing Systems 1997.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC