Learning with rethinking: Recurrently improving convolutional neural networks through feedback

作者：

Highlights：

• “Learning with Rethinking” algorithm is proposed to boost the perfor- mance of a baseline model.

• Feedback layer is proposed to propagate predicted probability to a bottom 5 layer.

• Different feature channels are weighted with emphasis layer according to top-down information.

摘要

•“Learning with Rethinking” algorithm is proposed to boost the perfor- mance of a baseline model.•Feedback layer is proposed to propagate predicted probability to a bottom 5 layer.•Different feature channels are weighted with emphasis layer according to top-down information.

论文关键词：Convolutional neural network,Image classification,Deep learning

论文评审过程：Received 7 September 2016, Revised 26 September 2017, Accepted 14 January 2018, Available online 31 January 2018, Version of Record 19 February 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.01.015