关键词: 主动学习 误分类采样 版本空间缩减

Abstract:

By selecting the most informative instances for labeling, active learners can significantly reduce the instance complexity of learning methods. We pre sent the sampling strategy of reducing the volume of version space by more than half while the leading active learning methods utilize the strategy of halving version spaces when sampling. Via the Halving Model, we propose an active learning method called CBMPMS（Committee-Based Most Possible Misclassif fication Sampiing） ,which samples the instances that have the largest probability to be misclassified by the current classifier. The method calculates  the entropy of the difference between the class probability distribution predicted by the current classifier ,and the committee selected randomly from  the current version space, and then takes it as the sampling criterion. The experiments on the UCI datasets show that the proposed method outperforms the traditional sampling methods on most selected datasets.

Key words: active learning, misclassification sampling, version sapee reduction