A Primal-Dual Convergence Analysis of Boosting
Abstract
Boosting combines weak learners into a predictor with low empirical risk. Its dual constructs a high entropy distribution upon which weak learners and training labels are uncorrelated. This manuscript studies this primal-dual relationship under a broad family of losses, including the exponential loss of AdaBoost and the logistic loss, revealing: - Weak learnability aids the whole loss family: for any {\epsilon}>0, O(ln(1/{\epsilon})) iterations suffice to produce a predictor with empirical risk {\epsilon}-close to the infimum; - The circumstances granting the existence of an empirical risk minimizer may be characterized in terms of the primal and dual problems, yielding a new proof of the known rate O(ln(1/{\epsilon})); - Arbitrary instances may be decomposed into the above two, granting rate O(1/{\epsilon}), with a matching lower bound provided for the logistic loss.
Cite
@article{arxiv.1101.4752,
title = {A Primal-Dual Convergence Analysis of Boosting},
author = {Matus Telgarsky},
journal= {arXiv preprint arXiv:1101.4752},
year = {2012}
}
Comments
40 pages, 8 figures; the NIPS 2011 submission "The Fast Convergence of Boosting" is a brief presentation of the primary results; compared with the JMLR version, this arXiv version has hyperref and some formatting tweaks