Today we saw how to obtain strong PAC learning algorithms from weak ones. We showed how to boost the accuracy using the Adaboost algorithm.
The first boosting algorithm is due to Rob Schapire and is described in his paper The Strength of Weak Learnability. Freund followed this work with a more practical boosting algorithm described in Boosting a Weak Learning Algorithm by Majority. Later on they combined their ideas into Adaboost, which was first described in the article A decision-theoretic generalization of on-line learning and an application to boosting (Freund and Schapire received the Gödel Prize for this work in 2003). There are many many *many* papers on boosting from the machine learning and statistics communities; more literature can be found in the page http://www.boosting.org.
Exercise 26: Re-do the proof for AdaBoost seen in class, completing all the details and the holes that were left out.
No comments:
Post a Comment