LOW: Training deep neural networks by learning optimal sample weights

作者：

Highlights：

• A new learning strategy is proposed, based on sample weighting.

• Sample weights are computed through an optimization problem that aims to maximize the decrease in the loss function.

• The proposed strategy allows deep neural network to achieve better generalization performances.

• The sample weights gives explainability to the training process by allowing an analysis of the most relevant samples.

摘要

•A new learning strategy is proposed, based on sample weighting.•Sample weights are computed through an optimization problem that aims to maximize the decrease in the loss function.•The proposed strategy allows deep neural network to achieve better generalization performances.•The sample weights gives explainability to the training process by allowing an analysis of the most relevant samples.

论文关键词：Deep learning,Sample weighting,Imbalanced data sets

论文评审过程：Received 15 July 2019, Revised 3 July 2020, Accepted 9 August 2020, Available online 12 August 2020, Version of Record 1 November 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107585