A deep convolutional neural network module that promotes competition of multiple-size filters

作者：

Highlights：

• A new deep ConvNet module that promotes competition amongst a set of multiple size filters.

• The competition is promoted by pooling the filter responses with max-pooling operation.

• This module can prevent filter co-adaptation.

• The results are competitive with the state-of-the-art results on MNIST, CIFAR-10, CIFAR-100, SVHN and ImageNet.

摘要

•A new deep ConvNet module that promotes competition amongst a set of multiple size filters.•The competition is promoted by pooling the filter responses with max-pooling operation.•This module can prevent filter co-adaptation.•The results are competitive with the state-of-the-art results on MNIST, CIFAR-10, CIFAR-100, SVHN and ImageNet.

论文关键词：Deep learning,Multi-size filter,Filter co-adaptation,Classification

论文评审过程：Received 23 December 2016, Revised 3 May 2017, Accepted 25 May 2017, Available online 31 May 2017, Version of Record 12 July 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.05.024