Spectral coefficients and classifier correlation

Authors:
Terry Windeatt;R. Ghaderi;G. Ardeshir
Affiliations:
Centre for Vision, Speech and Signal Processing, University of Surrey, Guildford, Surrey, UK;Centre for Vision, Speech and Signal Processing, University of Surrey, Guildford, Surrey, UK;Centre for Vision, Speech and Signal Processing, University of Surrey, Guildford, Surrey, UK
Venue:
MCS'03 Proceedings of the 4th international conference on Multiple classifier systems
Year:
2003

Citing 6
Cited 0

Spectral technique for hidden layer neural network training

Pattern Recognition Letters
Recursive Partitioning Technique for Combining Multiple Classifiers

Neural Processing Letters
Spectral Techniques in Digital Logic

Spectral Techniques in Digital Logic
Using Correspondence Analysis to Combine Classifiers

Machine Learning
Variance and Bias for General Loss Functions

Machine Learning
Boosted Tree Ensembles for Solving Multiclass Problems

MCS '02 Proceedings of the Third International Workshop on Multiple Classifier Systems

Quantified Score

Hi-index	0.00

Visualization

Abstract

Various counting measures, such as Margin and Bias/ Variance, have been proposed for analysing Multiple Classifier Systems (MCS) performance. In this paper a measure based on counting votes to estimate first order spectral coefficients for two-class problems is described. Experiments employing MLP base classifiers, in which parameters are fixed but systematically varied, demonstrate how the proposed measure varies with test error. Estimated spectral coefficients are used to design a weighted vote combiner, which is shown experimentally to be less sensitive than majority vote to base classifier complexity.