Regression with re-labeling for noisy data
作者：
Highlights：
• Novel active learning framework based on expectation-refinement sampling is presented.
• It performs active learning with re-labeling for regression task with noisy annotator.
• Exploration step selects unlabeled instance to be labeled next.
• Refinement step labels again for labeled instance to improve label accuracy.
• Experimental results demonstrate its effectiveness on benchmark datasets.
摘要
•Novel active learning framework based on expectation-refinement sampling is presented.•It performs active learning with re-labeling for regression task with noisy annotator.•Exploration step selects unlabeled instance to be labeled next.•Refinement step labels again for labeled instance to improve label accuracy.•Experimental results demonstrate its effectiveness on benchmark datasets.
论文关键词：Active learning,Re-labeling,Exploration-refinement sampling,Regression
论文评审过程：Received 13 October 2017, Revised 5 August 2018, Accepted 15 August 2018, Available online 17 August 2018, Version of Record 4 September 2018.
论文官网地址：https://doi.org/10.1016/j.eswa.2018.08.032
