Multi-granularity active learning based on the three-way decision

The reliance on data and the high cost of labelling are main problems facing deep learning today. Active aims to make best model with as few training samples possible. Previous query strategies for active have mainly used uncertainty diversity criteria, not considered distribution's multi-granularity. To extract more valid information from samples, we use three-way decisions select uncertain propose a multi-granularity method (MGAL). divides unlabeled into three parts: positive, negative boundary region. Through iterative decision delay domain can reduce cost. We validated five UCI datasets CIFAR10 dataset. experimental results show that is lower than two-way decisions. achieves good classification results, which validates model. In this case study, reader learn about ideas methods theory applied learning.