We present an extension of Freund and Schapire's AdaBoost algorithm that allows an input-dependent combination of the base hypotheses. A separate weak learner is used for determining the input-dependent weights of each hypothesis. The error function minimi ...
IDIAP1999