A 'No Panacea Theorem' for Multiple Classifier Combination

Authors:
Roland Hu;R. I. Damper
Affiliations:
University of Southampton Southampton SO17 1BJ, UK;University of Southampton Southampton SO17 1BJ, UK
Venue:
ICPR '06 Proceedings of the 18th International Conference on Pattern Recognition - Volume 02
Year:
2006

Citing 0
Cited 2

A 'No Panacea Theorem' for classifier combination

Pattern Recognition
Optimal weighting of bimodal biometric information with specific application to audio-visual person identification

Information Fusion

Quantified Score

Hi-index	0.00

Visualization

Abstract

We introduce the 'No Panacea Theorem' for classifier combination in the two-classifier, two-class case. It states that if the combination function is continuous and diverse, there exists a situation in which the combination algorithm will always give very bad performance. Thus, there is no optimal algorithm, suitable in all situations. From this theorem, we see that the probability density functions (pdf's) play an important role in the performance of combination algorithms, so studying the pdf's becomes the first step in finding a good algorithm.