We analyze the performance of boosting algorithms that make only a single pass over a collection of base classifiers. Our results indicate that ``picky'' boosters, which abstain from using base classifiers whose advantage is too small, can perform effectively in this setting.
Postscript or pdf.