Difficulty-weighted learning: A novel curriculum-like approach based on difficult examples for neural network training

作者：

Highlights：

• We prioritize the classification of difficult examples over easy examples.

• We proposed difficulty-weighted learning (DWL) for neural network training.

• DWL uses a loss function weighted by the neural network outputs.

• We evaluated the performance of DWL on several benchmark datasets.

• DWL has better generalization performance for MLP or a small CNN.

摘要

•We prioritize the classification of difficult examples over easy examples.•We proposed difficulty-weighted learning (DWL) for neural network training.•DWL uses a loss function weighted by the neural network outputs.•We evaluated the performance of DWL on several benchmark datasets.•DWL has better generalization performance for MLP or a small CNN.

论文关键词：Neural network,Curriculum learning,Supervised learning,Deep learning,Multilayer perceptron,Classification

论文评审过程：Received 20 December 2018, Revised 6 June 2019, Accepted 6 June 2019, Available online 8 June 2019, Version of Record 14 June 2019.

论文官网地址：https://doi.org/10.1016/j.eswa.2019.06.017