Filter Pruning via Learned Representation Median in the Frequency Domain

被引:9
|
作者
Zhang, Xin [1 ]
Xie, Weiying [1 ]
Li, Yunsong [1 ]
Lei, Jie [1 ]
Du, Qian [2 ]
机构
[1] Xidian Univ, State Key Lab Integrated Serv Networks, Xian 710071, Peoples R China
[2] Mississippi State Univ, Dept Elect & Comp Engn, Starkville, MS 39759 USA
基金
中国国家自然科学基金;
关键词
Frequency-domain analysis; Discrete cosine transforms; Convolution; Visual systems; Quantization (signal); Transforms; Redundancy; Absolutely unimportant filter pruning; frequency-domain transform; learned representation median (RM);
D O I
10.1109/TCYB.2021.3124284
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
In this article, we propose a novel filter pruning method for deep learning networks by calculating the learned representation median (RM) in frequency domain (LRMF). In contrast to the existing filter pruning methods that remove relatively unimportant filters in the spatial domain, our newly proposed approach emphasizes the removal of absolutely unimportant filters in the frequency domain. Through extensive experiments, we observed that the criterion for ``relative unimportance'' cannot be generalized well and that the discrete cosine transform (DCT) domain can eliminate redundancy and emphasize low-frequency representation, which is consistent with the human visual system. Based on these important observations, our LRMF calculates the learned RM in the frequency domain and removes its corresponding filter, since it is absolutely unimportant at each layer. Thanks to this, the time-consuming fine-tuning process is not required in LRMF. The results show that LRMF outperforms state-of-the-art pruning methods. For example, with ResNet110 on CIFAR-10, it achieves a 52.3% FLOPs reduction with an improvement of 0.04% in Top-1 accuracy. With VGG16 on CIFAR-100, it reduces FLOPs by 35.9% while increasing accuracy by 0.5%. On ImageNet, ResNet18 and ResNet50 are accelerated by 53.3% and 52.7% with only 1.76% and 0.8% accuracy loss, respectively. The code is based on PyTorch and is available at https://github.com/zhangxin-xd/LRMF.
引用
收藏
页码:3165 / 3175
页数:11
相关论文
共 50 条
  • [1] Filter Pruning via Geometric Median for Deep Convolutional Neural Networks Acceleration
    He, Yang
    Liu, Ping
    Wang, Ziwei
    Hu, Zhilan
    Yang, Yi
    [J]. 2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2019), 2019, : 4335 - 4344
  • [2] Filter pruning with uniqueness mechanism in the frequency domain for efficient neural networks
    Zhang, Shuo
    Gao, Mingqi
    Ni, Qiang
    Han, Jungong
    [J]. NEUROCOMPUTING, 2023, 530 : 116 - 124
  • [3] FILTER PRUNING VIA SOFTMAX ATTENTION
    Cho, Sungmin
    King, Hyeseong
    Kwon, Junseok
    [J]. 2021 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), 2021, : 3507 - 3511
  • [4] Pruning Hypothesis Spaces Using Learned Domain Theories
    Svatos, Martin
    Sourek, Gustav
    Zelezny, Filip
    Schockaert, Steven
    Kuzelka, Ondrej
    [J]. INDUCTIVE LOGIC PROGRAMMING (ILP 2017), 2018, 10759 : 152 - 168
  • [5] Median algorithms - Characterized in frequency domain
    Sawicki, J
    [J]. 2003 IEEE INTERNATIONAL SYMPOSIUM ON INTELLIGENT SIGNAL PROCESSING, PROCEEDINGS: FROM CLASSICAL MEASUREMENT TO COMPUTING WITH PERCEPTIONS, 2003, : 203 - 207
  • [6] Frequency domain median-like filter for periodic and quasi-periodic noise removal
    Aizenberg, I
    Butakoff, C
    [J]. IMAGE PROCESSING: ALGORITHMS AND SYSTEMS, 2002, 4667 : 181 - 191
  • [7] Filter pruning via expectation-maximization
    Xu, Sheng
    Li, Yanjing
    Yang, Linlin
    Zhang, Baochang
    Sun, Dianmin
    Liu, Kexin
    [J]. NEURAL COMPUTING & APPLICATIONS, 2022, 34 (15): : 12807 - 12818
  • [8] Filter pruning via expectation-maximization
    Sheng Xu
    Yanjing Li
    Linlin Yang
    Baochang Zhang
    Dianmin Sun
    Kexin Liu
    [J]. Neural Computing and Applications, 2022, 34 : 12807 - 12818
  • [9] Adaptive Filter Pruning via Sensitivity Feedback
    Zhang, Yuyao
    Freris, Nikolaos M.
    [J]. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2024, 35 (08) : 10996 - 11008
  • [10] Towards efficient filter pruning via topology
    Xiaozhou Xu
    Jun Chen
    Hongye Su
    Lei Xie
    [J]. Journal of Real-Time Image Processing, 2022, 19 : 639 - 649