A Method of Active Learning Based on Fuzzy Set Theory
Feng Hu, Lei Zhou, Jin Dai, Ke Liu · 2019
Active learning is one of the hotspots in the field of machine learning. The most crucial problem lies in the design of an appropriate valued criterion for selecting the most valuable unlabeled sample. Traditional active learning does not consider the fuzziness in the framework of active learning, which could significantly limit their performance. In this paper, an active learning method based on fuzzy set theory (FT-AL) is proposed to solve this problem. First, multi-criteria are established, which aim at the fuzziness in the framework of active learning. Second, weight factors are introduced into the multi-criteria. Finally, a fuzzy comprehensive evaluation model is established, which could evaluate the effect of the samples to be labeled. The experimental results show that the proposed methods have positive effects on improving the effect of base classifiers.