Semi-supervised learning combining co-training with active learning
作者：
Highlights：
• Our semi-supervised algorithm combines the benefits of both co-training and active learning.
• The most reliable instances are selected according to high confidence and nearest neighbor.
• We define contribution degree as the selection criteria of informative instances.
• Our algorithm achieves significant improvement for sacrificing same amount of human effort.
• Compared with standard co-training, our algorithm worked well on small labeled training sets.
摘要
•Our semi-supervised algorithm combines the benefits of both co-training and active learning.•The most reliable instances are selected according to high confidence and nearest neighbor.•We define contribution degree as the selection criteria of informative instances.•Our algorithm achieves significant improvement for sacrificing same amount of human effort.•Compared with standard co-training, our algorithm worked well on small labeled training sets.
论文关键词：Semi-supervised learning,Co-training,Confidence estimation,Active learning,Informative instances
论文评审过程：Available online 2 October 2013.
论文官网地址：https://doi.org/10.1016/j.eswa.2013.09.035
