nFormal framework for Active Learning in BNs
nEfficient
algorithm
nTheoretical gains supported by experimental results
n
nParticularly
useful when data is expensive
n
nMost
beneficial when domain
consists of distinctive common
and rare cases
nRandom
sampling does not look at rare cases enough
nIn medical
domains rare cases often even more important
nCan
adapt loss function to place even greater emphasis on them