Filters








27 Hits in 2.2 sec

MetaPruning: Meta Learning for Automatic Neural Network Channel Pruning [article]

Zechun Liu, Haoyuan Mu, Xiangyu Zhang, Zichao Guo, Xin Yang, Tim Kwang-Ting Cheng, Jian Sun
2019 arXiv   pre-print
In this paper, we propose a novel meta learning approach for automatic channel pruning of very deep neural networks.  ...  We first train a PruningNet, a kind of meta network, which is able to generate weight parameters for any pruned structure given the target network.  ...  In this section, we introduce our meta learning approach for automatically pruning channels in deep neural networks, that pruned network could meet various constraints easily.  ... 
arXiv:1903.10258v3 fatcat:7fp5bmuduzcapgtvtjjp6tkt2a

Graph Pruning for Model Compression [article]

Mingyang Zhang, Xinyi Yu, Jingtao Rong, Linlin Ou
2021 arXiv   pre-print
Subsequently, the best configuration of the Pruned Network is searched by reinforcement learning.  ...  Any series of the network is viewed as a graph. To automatically aggregate neighboring features for each node, a graph aggregator based on graph convolution networks(GCN) is designed.  ...  . • We combine Graph PruningNet with meta-learning for training and reinforcement learning for searching.  ... 
arXiv:1911.09817v2 fatcat:p6ivhwdtqrdrlgocu5tymxqoci

DHP: Differentiable Meta Pruning via HyperNetworks [article]

Yawei Li, Shuhang Gu, Kai Zhang, Luc Van Gool, Radu Timofte
2020 arXiv   pre-print
Network pruning has been the driving force for the acceleration of neural networks and the alleviation of model storage/transmission burden.  ...  To circumvent this problem, this paper introduces a differentiable pruning method via hypernetworks for automatic network pruning.  ...  .: MetaPruning: Meta learning for automatic neural network channel pruning. In: Proc.  ... 
arXiv:2003.13683v3 fatcat:2haocmohkzeklcpbmpruq45v4m

Evolving Transferable Pruning Functions [article]

Yuchen Liu, S.Y. Kung, David Wentzlaff
2021 arXiv   pre-print
Channel pruning has made major headway in the design of efficient deep learning models.  ...  Conventional approaches adopt human-made pruning functions to score channels' importance for channel pruning, which requires domain knowledge and could be sub-optimal.  ...  While these works use handcrafted scoring metrics, we learn transferable and generalizable pruning functions automatically. Meta-Learning.  ... 
arXiv:2110.10876v1 fatcat:buewobh6wffmzngs7iasn7wsty

Joint Multi-Dimension Pruning via Numerical Gradient Update [article]

Zechun Liu and Xiangyu Zhang and Zhiqiang Shen and Zhe Li and Yichen Wei and Kwang-Ting Cheng and Jian Sun
2021 arXiv   pre-print
We present joint multi-dimension pruning (abbreviated as JointPruning), an effective method of pruning a network on three crucial aspects: spatial, depth and channel simultaneously.  ...  size, depth) and construct a unique mapping from the pruning vector to the pruned network structures.  ...  or a meta network [37] can automatically decide the best pruning ratio.  ... 
arXiv:2005.08931v2 fatcat:yvnze4n7kzh43hnpj7emq5ye6u

Revisiting Parameter Sharing for Automatic Neural Channel Number Search

Jiaxing Wang, Haoli Bai, Jiaxiang Wu, Xupeng Shi, Junzhou Huang, Irwin King, Michael R. Lyu, Jian Cheng
2020 Neural Information Processing Systems  
Recent advances in neural architecture search inspire many channel number search algorithms (CNS) for convolutional neural networks.  ...  In this paper, we aim at providing a better understanding and exploitation of parameter sharing for CNS.  ...  For a L-layer neural network, layerwise channel number decisions are sampled from ⇡(✓), i.e. a = [a 1 , ..., a L ] ⇠ ⇡(✓), where a l 2 A = {1, 2, ..., A}, and A represents the index set of channel number  ... 
dblp:conf/nips/WangBWSHKL020 fatcat:oodjqieqqzbqxln6obpxvkxyoy

Differentiable Network Pruning via Polarization of Probabilistic Channelwise Soft Masks

Ming Ma, Jiapeng Wang, Zhenhua Yu, M. Hassaballah
2022 Computational Intelligence and Neuroscience  
Channel pruning has been demonstrated as a highly effective approach to compress large convolutional neural networks.  ...  For instance, our method prunes 65.91% FLOPs of ResNet50 on the ImageNet dataset with only 0.7% model accuracy degradation.  ...  [22] propose meta-attribute-based filter pruning (MFP), which adaptively selects the most appropriate pruning standard through an attribute (meta-attribute) of the current state of the neural network  ... 
doi:10.1155/2022/7775419 pmid:35571691 pmcid:PMC9098282 fatcat:r4kalfper5bdfhcxhqc63td5pe

Discrimination-aware Network Pruning for Deep Model Compression [article]

Jing Liu, Bohan Zhuang, Zhuangwei Zhuang, Yong Guo, Junzhou Huang, Jinhui Zhu, Mingkui Tan
2020 arXiv   pre-print
We study network pruning which aims to remove redundant channels/kernels and hence speed up the inference of deep networks.  ...  To prevent DCP/DKP from selecting redundant channels/kernels, we propose a new adaptive stopping condition, which helps to automatically determine the number of selected channels/kernels and often results  ...  Apart from these methods, the pruning ratio for each layer can also be automatically determined by reinforcement learning [25] , [71] or meta-learning [48] .  ... 
arXiv:2001.01050v1 fatcat:vdy4g473czcdphd2wyqegiboyu

Dynamic Slimmable Network [article]

Changlin Li, Guangrun Wang, Bing Wang, Xiaodan Liang, Zhihui Li, Xiaojun Chang
2021 arXiv   pre-print
Current dynamic networks and dynamic pruning methods have shown their promising capability in reducing theoretical computation complexity.  ...  In the first stage, a novel training technique for weight-sharing networks named In-place Ensemble Bootstrapping is proposed to improve the supernet training efficacy.  ...  Remarkably, DS-MBNet outperforms the SOTA pruning methods EagleEye [40] and Meta-Pruning [51] by 1.9% and 2.2%.  ... 
arXiv:2103.13258v1 fatcat:q3dqpaakf5bilikvsn7akzhone

Building Efficient CNNs Using Depthwise Convolutional Eigen-Filters (DeCEF) [article]

Yinan Yu and Samuel Scheidegger and Tomas McKelvey
2022 arXiv   pre-print
Deep Convolutional Neural Networks (CNNs) have been widely used in various domains due to their impressive capabilities.  ...  To reduce the complexity of a network, compression techniques can be applied. These methods typically rely on the analysis of trained deep learning models.  ...  [40] proposes a meta network, which is able to generate weight parameters for any pruned structure given the target network, which can be used to search for goodperforming pruned networks. [58] introduce  ... 
arXiv:1910.09359v3 fatcat:frtr2r52x5bfvpca44s337eiiy

SNF: Filter Pruning via Searching the Proper Number of Filters [article]

Pengkun Liu, Yaru Yue, Yanjun Guo, Xingxiang Tao, Xiaoguang Zhou
2021 arXiv   pre-print
neural networks' redundancy.  ...  Convolutional Neural Network (CNN) has an amount of parameter redundancy, filter pruning aims to remove the redundant filters and provides the possibility for the application of CNN on terminal devices  ...  Metapruning [36] utilized meta-learning to train a pruning network which provides weights for all the possible sub-networks, and then searched for the best pruned network structures. Pavlo et al.  ... 
arXiv:2112.07282v1 fatcat:4p3trr6i6nhobbjbdtyak3g3vi

ResRep: Lossless CNN Pruning via Decoupling Remembering and Forgetting [article]

Xiaohan Ding, Tianxiang Hao, Jianchao Tan, Ji Liu, Jungong Han, Yuchen Guo, Guiguang Ding
2021 arXiv   pre-print
We propose ResRep, a novel method for lossless channel pruning (a.k.a. filter pruning), which slims down a CNN by reducing the width (number of output channels) of convolutional layers.  ...  maintain the performance and the latter learn to prune.  ...  biological neural network, which resembles pruning.  ... 
arXiv:2007.03260v4 fatcat:6pyaguorqfgexmvttzzp6wxhre

Adjoined Networks: A Training Paradigm with Applications to Network Compression [article]

Utkarsh Nath, Shrinu Kushagra, Yingzhen Yang
2022 arXiv   pre-print
We further propose Differentiable Adjoined Networks (DAN), a training paradigm that augments AN by using neural architecture search to jointly learn both the width and the weights for each layer of the  ...  In this paper, we introduce Adjoined Networks, or AN, a learning paradigm that trains both the original base network and the smaller compressed network together.  ...  A recent study Meta Pruning [29] searches over the number of channels in each layer. It generates weights for all candidates and then selects the architecture with the highest validation accuracy.  ... 
arXiv:2006.05624v5 fatcat:chp5hevvbrdvlc65qpipa4xkgy

DS-Net++: Dynamic Weight Slicing for Efficient Inference in CNNs and Transformers [article]

Changlin Li, Guangrun Wang, Bing Wang, Xiaodan Liang, Zhihui Li, Xiaojun Chang
2021 arXiv   pre-print
Here, we explore a hardware-efficient dynamic inference regime, named dynamic weight slicing, which adaptively slice a part of network parameters for inputs with diverse difficulty levels, while keeping  ...  Based on this scheme, we present dynamic slimmable network (DS-Net) and dynamic slice-able network (DS-Net++) by input-dependently adjusting filter numbers of CNNs and multiple dimensions in both CNNs  ...  Channel Gating Neural Network [26] and FBS [27] identify and skip the unimportant input channels at run-time.  ... 
arXiv:2109.10060v1 fatcat:s5nmhmobsjc7tkstsiwqj5qesu

Provable Filter Pruning for Efficient Neural Networks [article]

Lucas Liebenwein, Cenk Baykal, Harry Lang, Dan Feldman, Daniela Rus
2020 arXiv   pre-print
We present a provable, sampling-based approach for generating compact Convolutional Neural Networks (CNNs) by identifying and removing redundant filters from an over-parameterized network.  ...  In contrast to existing filter pruning approaches, our method is simultaneously data-informed, exhibits provable guarantees on the size and performance of the pruned network, and is widely applicable to  ...  Metapruning: Meta learning for automatic neural network channel pruning. In Proceedings of the IEEE International Conference on Computer Vision, pp. 3296-3305, 2019a.  ... 
arXiv:1911.07412v2 fatcat:l5drcoblgvdxfcksho5g7inhue
« Previous Showing results 1 — 15 out of 27 results