1: \begin{abstract}

2: % This is a great paper and it has a concise abstract. 

3:   The significance of the study of the theoretical and practical

4:   properties of AdaBoost is unquestionable, given its simplicity, wide

5:   practical use, and effectiveness on real-world datasets. Here we

6:   present a few open problems regarding the behavior of ``Optimal

7:   AdaBoost,'' a term coined by Rudin, Daubechies, and Schapire in 2004

8:   to label the simple version of the standard AdaBoost algorithm in

9:   which the weak learner that AdaBoost uses always outputs

10:   the weak classifier with lowest weighted error among the respective

11:   hypothesis class of weak classifiers implicit in the weak learner.

12:   We concentrate on the standard, ``vanilla'' version of Optimal

13:   AdaBoost for binary classification that results from using an

14:   exponential-loss upper bound on the misclassification training

15:   error. We present two types of open problems. One deals with general

16:   weak hypotheses. The other deals with the particular case of

17:   decision stumps, as often and commonly used in practice. Answers to

18:   the open problems can have immediate significant impact to

19:   (1) cementing previously established results on asymptotic convergence properties of Optimal AdaBoost, for finite datasets,

20:   which in turn can be the start to any convergence-rate analysis; (2)

21:   understanding the weak-hypotheses class of effective decision stumps generated

22:   from data, which we have empirically observed to be significantly smaller than

23:   the typically obtained class, as well as the

24:   effect on the weak learner's running time and previously established

25:   improved bounds on the generalization performance of Optimal AdaBoost

26:   classifiers; and (3) shedding some light on the ``self control'' that

27:   AdaBoost tends to exhibit in practice.

28: \end{abstract}

29: