It is possible to employ more base learners for disagreement-based semisupervised
learning as tritraining [Zhou and Li 2005b] and coforest [Li and Zhou 2007] did, and
for active learning as query-by-committee [Seung et al. 1992] did. However, the combination
of such kind of semisupervised learning and active learning techniques will
encounter a big challenge, that is, how to maintain the diversity among the learners in
the learning process, as it is well known that the diversity is crucial for disagreementbased
approaches to succeed [Wang and Zhou 2007]. Thus, our approach only uses two
base learners, and such a process has a sound theoretical support [Wang and Zhou
2008].