On Implicit Filter Level Sparsity in Convolutional Neural Networks

We investigate filter level sparsity that emerges in convolutional neural networks (CNNs) which employ Batch Normalization and ReLU activation, and are trained with adaptive gradient descent techniques and L2 regularization or weight decay. We conduct an extensive experimental study casting our initial findings into hypotheses and conclusions about the mechanisms underlying the emergent filter level sparsity. This study allows new insight into the performance gap obeserved between adapative and non-adaptive gradient descent methods in practice. Further, analysis of the effect of training strategies and hyperparameters on the sparsity leads to practical suggestions in designing CNN training strategies enabling us to explore the tradeoffs between feature selectivity, network capacity, and generalization performance. Lastly, we show that the implicit sparsity can be harnessed for neural network speedup at par or better than explicit sparsification / pruning approaches, with no modifications to the typical training pipeline required.

[1]  Song Han,et al.  AMC: AutoML for Model Compression and Acceleration on Mobile Devices , 2018, ECCV.

[2]  Li Fei-Fei,et al.  ImageNet: A large-scale hierarchical image database , 2009, CVPR.

[3]  Luca Antiga,et al.  Automatic differentiation in PyTorch , 2017 .

[4]  Matthew Botvinick,et al.  On the importance of single directions for generalization , 2018, ICLR.

[5]  Timo Aila,et al.  Pruning Convolutional Neural Networks for Resource Efficient Inference , 2016, ICLR.

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

[7]  Jorge Nocedal,et al.  On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima , 2016, ICLR.

[8]  Gregory J. Wolff,et al.  Optimal Brain Surgeon and general network pruning , 1993, IEEE International Conference on Neural Networks.

[9]  Richard Socher,et al.  Improving Generalization Performance by Switching from Adam to SGD , 2017, ArXiv.

[10]  Lucas Theis,et al.  Faster gaze prediction with dense networks and Fisher pruning , 2018, ArXiv.

[11]  Taiji Suzuki,et al.  Adam Induces Implicit Weight Sparsity in Rectifier Neural Networks , 2018, 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA).

[12]  Leonidas J. Guibas,et al.  ObjectNet3D: A Large Scale Database for 3D Object Recognition , 2016, ECCV.

[13]  Bolei Zhou,et al.  Revisiting the Importance of Individual Units in CNNs via Ablation , 2018, ArXiv.

[14]  Razvan Pascanu,et al.  Sharp Minima Can Generalize For Deep Nets , 2017, ICML.

[15]  Song Han,et al.  Deep Compression: Compressing Deep Neural Network with Pruning, Trained Quantization and Huffman Coding , 2015, ICLR.

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

[17]  Yoram Singer,et al.  Adaptive Subgradient Methods for Online Learning and Stochastic Optimization , 2011, J. Mach. Learn. Res..

[18]  Frank Hutter,et al.  Decoupled Weight Decay Regularization , 2017, ICLR.

[19]  James Zijun Wang,et al.  Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers , 2018, ICLR.

[20]  Mingjie Sun,et al.  Rethinking the Value of Network Pruning , 2018, ICLR.

[21]  R. Venkatesh Babu,et al.  Data-free Parameter Pruning for Deep Neural Networks , 2015, BMVC.

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

[23]  Zhiqiang Shen,et al.  Learning Efficient Convolutional Networks through Network Slimming , 2017, 2017 IEEE International Conference on Computer Vision (ICCV).

[24]  Frank Hutter,et al.  Fixing Weight Decay Regularization in Adam , 2017, ArXiv.

[25]  Michael C. Mozer,et al.  Skeletonization: A Technique for Trimming the Fat from a Network via Relevance Assessment , 1988, NIPS.

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

[27]  Claus Nebauer,et al.  Evaluation of convolutional neural networks for visual recognition , 1998, IEEE Trans. Neural Networks.

[28]  Yiran Chen,et al.  Learning Structured Sparsity in Deep Neural Networks , 2016, NIPS.

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

[30]  Joseph Paul Cohen,et al.  RandomOut: Using a convolutional gradient norm to rescue convolutional filters , 2016, 1602.05931.

[31]  Sanjiv Kumar,et al.  On the Convergence of Adam and Beyond , 2018 .

[32]  Matthew D. Zeiler ADADELTA: An Adaptive Learning Rate Method , 2012, ArXiv.

[33]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[34]  Alex Krizhevsky,et al.  Learning Multiple Layers of Features from Tiny Images , 2009 .

[35]  Rui Peng,et al.  Network Trimming: A Data-Driven Neuron Pruning Approach towards Efficient Deep Architectures , 2016, ArXiv.

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