Boosting Revisited

In this paper some boosting and on-line allocation strategies are revisited in order to provide new proofs inspired by Chernoo's bounding techniques taken from common Probability Theory. Even if some of the results presented here are weaker than the best known today, the generalized style of the proofs hopefully ooers a better understanding of known results and suggests new strategies. This paper is a Minor Thesis for the PhD program of \Universitt a Statale di Milano". 1 Intorduction The topics treated in this paper can be informally summarized as follows: On-line allocation: for a certain number of rounds a xed amount of a given resource is to be allocated among a set of possible choices judging only by the outcomes of previous rounds. Show that, even ignoring any a priori knowledge about the performance of each individual choice, an allocation strategy exists which never performs much worse than the best choice. PhD student at the \Universitt a Statale di Milano".