GridMix: Strong regularization through local context mapping

作者：

Highlights：

• Our GridMix prevents the model from only focusing on a few subregions through local context mapping.

• GridMix achieved state-of-the-art performances in classification (3 datasets) and robustness (1 dataset).

• Our analysis shows that local context mapping can effectively regularize the network model.

摘要

•Our GridMix prevents the model from only focusing on a few subregions through local context mapping.•GridMix achieved state-of-the-art performances in classification (3 datasets) and robustness (1 dataset).•Our analysis shows that local context mapping can effectively regularize the network model.

论文关键词：Deep learning,Network regularization,Data augmentation

论文评审过程：Received 10 May 2020, Revised 11 August 2020, Accepted 11 August 2020, Available online 14 August 2020, Version of Record 16 August 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107594