Convolutional neural network with nonlinear competitive units

作者：

Highlights：

•

• A novel nonlinear unit is proposed, named Nonlinear Competitive Unit, which can be regarded as a feature fusion method as well as an activation function.

• The convergence speed of our NCU-based model is improved accompanied with higher stability.

• The experiments validate that NCU-based models can effectively boost the performance in both face verification task and visual classification task.

摘要

•A novel nonlinear unit is proposed, named Nonlinear Competitive Unit, which can be regarded as a feature fusion method as well as an activation function.•The convergence speed of our NCU-based model is improved accompanied with higher stability.•The experiments validate that NCU-based models can effectively boost the performance in both face verification task and visual classification task.

论文关键词：Nonlinear competitive unit,Feature fusion,Activation function,Face verification,Visual classification

论文评审过程：Received 2 March 2017, Revised 25 August 2017, Accepted 28 September 2017, Available online 21 October 2017, Version of Record 3 November 2017.

论文官网地址：https://doi.org/10.1016/j.image.2017.09.011