TBAL: Two-stage batch-mode active learning for image classification

作者：

Highlights：

• A novel clustering based active learning method that can achieve better balance between uncertainty and diversity.

• Better robustness evaluated in a newly designed metric: transferred-supervised accuracy.

• Comparison with two strong semi-supervised learning based state-of-the-art active learning methods (VAAL and LL4AL) in two tasks (semi-supervise accuracy and transferred-supervised accuracy).

摘要

•A novel clustering based active learning method that can achieve better balance between uncertainty and diversity.•Better robustness evaluated in a newly designed metric: transferred-supervised accuracy.•Comparison with two strong semi-supervised learning based state-of-the-art active learning methods (VAAL and LL4AL) in two tasks (semi-supervise accuracy and transferred-supervised accuracy).

论文关键词：Active learning,Image classification,Semi-supervised learning

论文评审过程：Received 12 May 2020, Revised 24 June 2021, Accepted 2 May 2022, Available online 19 May 2022, Version of Record 6 June 2022.

论文官网地址：https://doi.org/10.1016/j.image.2022.116731