Contrastive learning from label distribution: A case study on text classification
Neurocomputing(2022)
摘要
•The proposed method learns under the supervision of the predicted label distribution.•The two contrastive losses work in the label space instead of the embedding space.•It leads to significant improvements over large label spaces or limited label data.•It captures label correlations better than working in the embedding space.
更多查看译文
关键词
Text classification,Contrastive learning,Label distribution,Deep neural network
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要