In order to deploy deep Convolutional Neural Networks (CNNs) on the mobile devices, many mobile CNNs are introduced. Currently, some online applications are usually re-trained because of the constantly-increasing data. However, compared with the regular models, it is not very efficient to train the present mobile models. Therefore, the purpose of this paper is to propose efficient mobile models both in the training and test processes through exploring the main causes of the current mobile CNNs’ inefficiency and the parameters’ properties. Finally, this paper introduces Highly Shared Convolutional Neural Networks (HSC-Nets). The HSC-Nets employ two shared mechanisms to reuse the filters comprehensively. Experimental results showed that, compared with the regular networks and the latest state-of-the-art group-conv mobile networks, the HSC-Nets can achieve promising performances and effectively decrease the model size. Furthermore, it is also more efficient in both the training and test processes.

为了在移动设备上部署深度卷积神经网络（CNN），引入了许多移动CNN。当前，由于数据的不断增加，通常会对一些在线应用程序进行重新培训。但是，与常规模型相比，训练当前的移动模型不是很有效。因此，本文的目的是通过探讨当前移动CNN效率低下的主要原因和参数的属性，在训练和测试过程中提出有效的移动模型。最后，本文介绍了高度共享的卷积神经网络（HSC-Nets）。HSC网络采用两种共享机制来全面重用过滤器。实验结果表明，与常规网络和最新的最先进的群组转换移动网络相比，HSC-Nets可以取得令人鼓舞的性能，并有效减小模型尺寸。此外，它在培训和测试过程中也更加有效。