A new perspective on boosting in linear regression via subgradient optimization and relatives (Q682283)

Boosting methods in statistical estimation, as e.g. in linear regression and classification, being linear combinations of simple estimators, or stage-wise procedures, have been interpreted by gradient-descent-type algorithms in some function spaces. Here, representations of boosting methods are given by means of subgradient descent procedures minimizing a certain maximum absolute correlation function. Moreover, some modifications of boosting methods and their convergence properties are provided. Numerical examples are given.