ADANOISE: Training neural networks with adaptive noise for imbalanced data classification
作者：
Highlights：
• We propose a novel training method based on adaptive noise for neural networks.
• Each minority instance is oversampled by adding different random noise vectors.
• The neural network and the parameter of the noise distribution are simultaneously trained.
• The noise distribution is adaptively determined toward improving the performance.
• Experimental results demonstrate its effectiveness under class imbalance.
摘要
•We propose a novel training method based on adaptive noise for neural networks.•Each minority instance is oversampled by adding different random noise vectors.•The neural network and the parameter of the noise distribution are simultaneously trained.•The noise distribution is adaptively determined toward improving the performance.•Experimental results demonstrate its effectiveness under class imbalance.
论文关键词：Neural network,Binary classification,Class imbalance,Adaptive noise
论文评审过程：Received 26 April 2021, Revised 7 November 2021, Accepted 30 November 2021, Available online 20 December 2021, Version of Record 24 December 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.116364
