In AdaBoost, we define the error for a base model fmx as m = yn,fmxn m n . We normally have m . We

In AdaBoost, we define the error for a base model fm¹xº as m =

Í

yn,fm¹xnº ¯¹mº

n . We normally have m

.

We then reweight the training samples for the next round as

Compute the error of the same base model fm¹xº on the reweighted data, that is,

and explain how ˜ m differs from the m+1 that will be computed in the next round.

-(m+1) (m) an e-yn Wmfm (xn) Vn = 1, 2,...,N. N (m) -n=1 e-yn Wmfm(xn)