(a) Recall the description of the AdaBoost algorithm. Prove that if the "wealk (t+1) w(t). Ifft classifies Xj incor- learner" ft classifies Xj correctly, then w rectly, then wt+1) w(t) (b) We will say that the class G of binary classifiers G O g:S->{1) satisfies the -weak learnability condition for the training data (X,K) E S {1}, 1, . . . , n, if for any collection of weights wl, . . . , wn 0, wj-1, there exists gE G such that for some 0 Prove that the following "margin" condition is sufficient for -weak learnabil- ity: there exist a1,... , a 2 0, (both depending on the given training data) such that for every i, 1 Si<>