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