Boosting-type algorithms, e.g. AdaBoost or Arc-GV are particular iterative strategies to minimize a constrained objective function – similar to Barrier algorithms. Based on this new understanding it is shown that convergence of Boosting becomes simpler to prove and we outline directions to develop further Boosting schemes. In particular a very natural Boosting technique for regression – Eps-Boost – is proposed.

This paper reduced my Erdös-number to three (Warmuth,Jagota,Erdös)