No-Regret Boosting

Authors:
Anna Gambin;Ewa Szczurek
Affiliations:
Inst. of Informatics, Warsaw University, Banacha 2, 02-097 Warsaw, Poland;Inst. of Informatics, Warsaw University, Banacha 2, 02-097 Warsaw, Poland and Max Planck Inst. for Molecular Genetics, Ihnestrasse 73, 14195 Berlin, Germany
Venue:
ICANNGA '07 Proceedings of the 8th international conference on Adaptive and Natural Computing Algorithms, Part I
Year:
2007

Citing 5
Cited 0

The weighted majority algorithm

Information and Computation
Game theory, on-line prediction and boosting

COLT '96 Proceedings of the ninth annual conference on Computational learning theory
Prediction games and arcing algorithms

Neural Computation
Potential-Based Algorithms in On-Line Prediction and Game Theory

Machine Learning
Sample classification from protein mass spectrometry, by 'peak probability contrasts'

Bioinformatics

Quantified Score

Hi-index	0.00

Visualization

Abstract

Following [4], we analyze boosting from a game-theoretic perspective. We define a wide class of boosting classification algorithms called H-boosting methods, which are based on Hannan-consistent game playing strategies. These strategies tend to minimize the regret of a player, i.e. are able to minimize the difference between its expected cumulative loss and the cumulative loss achievable using the single best strategy. The "weighted majority" boosting algorithm [4] is proved to belong to the class of H-boosting procedures. A new boosting algorithm is proposed, as an another example of such a regret-minimizing method.