The paper is organized as follows. In Section 2 we review ADTrees and the LogitBoost procedure. Section 3 describes our attempts to cast ADTrees to the multiclass setting. Section 4 describes the new algorithm that induces ADTrees
using LogitBoost. Section 5 contains experimental results that compare both the LogitBoost and AdaBoost.MH methods with the best of the adaptations of the original algorithm on some benchmark datasets. Section 6 summarizes the contributions made in this paper.