Committee-based algorithm
n
Train k classifiers C1, C2,.. Ck on training
data
n
For each unlabeled instance x
n
Find prediction y1,..,
yk from the k classifiers
n
Compute uncertainty U(x) as entropy of
above y-s
n
Pick instance with highest uncertainty
n
Sampling for representativeness:
n
With weight as U(x), do weighted sampling
to select an instance for labeling.