Multipath Ensemble Convolutional Neural Network

Convolutional neural networks (CNNs) have achieved great success in the field of computer vision in recent years. In order to achieve higher recognition accuracy, the layers of CNNs are generally required to be deeper. However, the increase of CNNs depth will lead to gradient vanishing, which makes insufficient training of network parameters that are close to the input layer and further results in performance degradation. In addition, the simple deepening of layers has limited effects on classification accuracy improvement. Compared with shallow network, the deep network is inherently difficult to be optimized, and thus, converges slowly. To this end, a novel multipath ensemble CNN (ME-CNN) model is proposed in this paper. On one hand, the ME-CNN directly concatenates the low-level with that of high-level output features, and thus, the error gradient can be back propagated through a shorter path to achieve the ultradeep network training. On another, the ME-CNN broadens the network width by increasing the number of feature channels to further improve the network performance. Experimental results on CIFAR and tiny ImageNet datasets verify that the depth of ME-CNN can reach beyond one hundred layers, and the performance gradually increases as the network deepens and widens.

[1]  Andrew Zisserman,et al.  Very Deep Convolutional Networks for Large-Scale Image Recognition , 2014, ICLR.

[2]  Andrew L. Maas Rectifier Nonlinearities Improve Neural Network Acoustic Models , 2013 .

[3]  Saeid Nahavandi,et al.  Multi-Residual Networks: Improving the Speed and Accuracy of Residual Networks , 2016, 1609.05672.

[4]  Sergey Ioffe,et al.  Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning , 2016, AAAI.

[5]  Qi Shi,et al.  A Deep Learning Approach to Network Intrusion Detection , 2018, IEEE Transactions on Emerging Topics in Computational Intelligence.

[6]  Geoffrey E. Hinton,et al.  On the importance of initialization and momentum in deep learning , 2013, ICML.

[7]  Ting Zhang,et al.  IGCV$2$: Interleaved Structured Sparse Convolutional Neural Networks , 2018 .

[8]  Parma Nand,et al.  Video Dynamics Detection Using Deep Neural Networks , 2018, IEEE Transactions on Emerging Topics in Computational Intelligence.

[9]  Geoffrey E. Hinton,et al.  Rectified Linear Units Improve Restricted Boltzmann Machines , 2010, ICML.

[10]  Zhuowen Tu,et al.  Deeply-Supervised Nets , 2014, AISTATS.

[11]  Sergey Ioffe,et al.  Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , 2015, ICML.

[12]  Enhua Wu,et al.  Squeeze-and-Excitation Networks , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[13]  Jian-Huang Lai,et al.  IGCV2: Interleaved Structured Sparse Convolutional Neural Networks , 2018, ArXiv.

[14]  Serge J. Belongie,et al.  Residual Networks Behave Like Ensembles of Relatively Shallow Networks , 2016, NIPS.

[15]  Tianqi Chen,et al.  Empirical Evaluation of Rectified Activations in Convolutional Network , 2015, ArXiv.

[16]  Qiang Chen,et al.  Network In Network , 2013, ICLR.

[17]  Xiaogang Wang,et al.  Residual Attention Network for Image Classification , 2017, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[18]  David A. Forsyth,et al.  Swapout: Learning an ensemble of deep architectures , 2016, NIPS.

[19]  Gregory Shakhnarovich,et al.  FractalNet: Ultra-Deep Neural Networks without Residuals , 2016, ICLR.

[20]  Kilian Q. Weinberger,et al.  Densely Connected Convolutional Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[21]  Zhuowen Tu,et al.  On the Connection of Deep Fusion to Ensembling , 2016, ArXiv.

[22]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[23]  Junmo Kim,et al.  Deep Pyramidal Residual Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[24]  Nitish Srivastava,et al.  Dropout: a simple way to prevent neural networks from overfitting , 2014, J. Mach. Learn. Res..

[25]  Gustavo Carneiro,et al.  Competitive Multi-scale Convolution , 2015, ArXiv.

[26]  Yoshua Bengio,et al.  Understanding the difficulty of training deep feedforward neural networks , 2010, AISTATS.

[27]  Jian Sun,et al.  Identity Mappings in Deep Residual Networks , 2016, ECCV.

[28]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[29]  Yoshua Bengio,et al.  Maxout Networks , 2013, ICML.

[30]  Wenjun Zeng,et al.  Deeply-Fused Nets , 2016, ArXiv.

[31]  Sepp Hochreiter,et al.  Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) , 2015, ICLR.

[32]  Trevor Darrell,et al.  Caffe: Convolutional Architecture for Fast Feature Embedding , 2014, ACM Multimedia.

[33]  Jian Sun,et al.  Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification , 2015, 2015 IEEE International Conference on Computer Vision (ICCV).

[34]  Jürgen Schmidhuber,et al.  Training Very Deep Networks , 2015, NIPS.

[35]  Nikos Komodakis,et al.  Wide Residual Networks , 2016, BMVC.

[36]  Diogo Almeida,et al.  Resnet in Resnet: Generalizing Residual Architectures , 2016, ArXiv.

[37]  Dumitru Erhan,et al.  Going deeper with convolutions , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[38]  Huanhuan Chen,et al.  A Cluster-Based Semisupervised Ensemble for Multiclass Classification , 2017, IEEE Transactions on Emerging Topics in Computational Intelligence.

[39]  Kilian Q. Weinberger,et al.  Deep Networks with Stochastic Depth , 2016, ECCV.