Part of Advances in Neural Information Processing Systems 12 (NIPS 1999)
Nigel Duffy, David Helmbold
Recent interpretations of the Adaboost algorithm view it as per(cid:173) forming a gradient descent on a potential function. Simply chang(cid:173) ing the potential function allows one to create new algorithms re(cid:173) lated to AdaBoost. However, these new algorithms are generally not known to have the formal boosting property. This paper ex(cid:173) amines the question of which potential functions lead to new al(cid:173) gorithms that are boosters. The two main results are general sets of conditions on the potential; one set implies that the resulting algorithm is a booster, while the other implies that the algorithm is not. These conditions are applied to previously studied potential functions , such as those used by LogitBoost and Doom II.