Highly shared Convolutional Neural Networks

作者：

Highlights：

• We explore the factors of CNNs’ inefficiency and the parameters’ properties.

• We propose mobile Highly Shared Convolutional Neural Networks (HSC-Nets).

• HSC-Nets share the parameters thoroughly by employing two novel shared methods.

• HSC-Nets can effectively improve efficiency both in the training and test.

摘要

•We explore the factors of CNNs’ inefficiency and the parameters’ properties.•We propose mobile Highly Shared Convolutional Neural Networks (HSC-Nets).•HSC-Nets share the parameters thoroughly by employing two novel shared methods.•HSC-Nets can effectively improve efficiency both in the training and test.

论文关键词：Deep learning,CNNs,Group convolutions,Highly shared convolutions,HSC-Nets

论文评审过程：Received 24 June 2019, Revised 13 January 2021, Accepted 22 February 2021, Available online 4 March 2021, Version of Record 19 March 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.114782