论文信息 - Pruning the Convolution Neural Network (SqueezeNet) based on L2 Normalization of Activation Maps

Pruning the Convolution Neural Network (SqueezeNet) based on L2 Normalization of Activation Maps

In recent years, deep learning models have become popular in real-time embedded application, but there are many complexities for the hardware deployment because of limited resources such as memory, computational power, and energy. Network pruning is one of the promising technique to solve these problems. This paper presents an approach to prune the Convolution Neural Network (SqueezeNet) based on L2 nor-malization of activation maps on the CIFAR-10 dataset without introducing the network sparsity in the pruned model. Results show that proposed approach reduces the SqueezeNet model by 67% without a significant drop in the accuracy of the model (optimal pruning efficiency result)

Mohamed El-Sharkawy | Akash Sunil Gaikwad

[1] Hanan Samet,et al. Pruning Filters for Efficient ConvNets , 2016, ICLR.

[2] Forrest N. Iandola,et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size , 2016, ArXiv.

[3] Jacek M. Zurada,et al. Building Efficient ConvNets using Redundant Feature Pruning , 2018, ArXiv.

[4] Andrew Lavin,et al. Fast Algorithms for Convolutional Neural Networks , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[5] Timo Aila,et al. Pruning Convolutional Neural Networks for Resource Efficient Transfer Learning , 2016, ArXiv.

[6] Michael I. Jordan,et al. Advances in Neural Information Processing Systems 30 , 1995 .

[7] Suvrit Sra,et al. Diversity Networks: Neural Network Compression Using Determinantal Point Processes , 2015, 1511.05077.

[8] Yann LeCun,et al. Optimal Brain Damage , 1989, NIPS.

[9] Wonyong Sung,et al. Structured Pruning of Deep Convolutional Neural Networks , 2015, ACM J. Emerg. Technol. Comput. Syst..

[10] Lior Wolf,et al. Channel-Level Acceleration of Deep Face Representations , 2015, IEEE Access.

[11] Yann LeCun,et al. Fast Training of Convolutional Networks through FFTs , 2013, ICLR.