Stacked BNAS: Rethinking Broad Convolutional Neural Network for Neural Architecture Search

被引:2
|
作者
Ding, Zixiang [1 ,2 ]
Chen, Yaran [1 ,2 ]
Li, Nannan [1 ,2 ]
Zhao, Dongbin [1 ,2 ]
Chen, C. L. Philip [3 ,4 ]
机构
[1] Chinese Acad Sci, Inst Automat, State Key Lab Multimodal Artificial Intelligence S, Beijing 100190, Peoples R China
[2] Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 100049, Peoples R China
[3] South China Univ Technol, Sch Comp Sci & Engn, Guangzhou 510641, Peoples R China
[4] Pazhou Lab, Ctr Affect Comp & Gen Models, Guangzhou 510335, Peoples R China
基金
中国国家自然科学基金;
关键词
Broad neural architecture search (BNAS); knowledge embedding search (KES); stacked broad convolutional neural network (BCNN); LEARNING-SYSTEM; APPROXIMATION;
D O I
10.1109/TSMC.2023.3275128
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Different from other deep scalable architecture-based NAS approaches, Broad Neural Architecture Search (BNAS) proposes a broad scalable architecture which consists of convolution and enhancement blocks, dubbed Broad Convolutional Neural Network (BCNN), as the search space for amazing efficiency improvement. BCNN reuses the topologies of cells in the convolution block so that BNAS can employ few cells for efficient search. Moreover, multi-scale feature fusion and knowledge embedding are proposed to improve the performance of BCNN with shallow topology. However, BNAS suffers some drawbacks: 1) insufficient representation diversity for feature fusion and enhancement and 2) time consumption of knowledge embedding design by human experts. This paper proposes Stacked BNAS, whose search space is a developed broad scalable architecture named Stacked BCNN, with better performance than BNAS. On the one hand, Stacked BCNN treats mini BCNN as a basic block to preserve comprehensive representation and deliver powerful feature extraction ability. For multi-scale feature enhancement, each mini BCNN feeds the outputs of deep and broad cells to the enhancement cell. For multi-scale feature fusion, each mini BCNN feeds the outputs of deep, broad and enhancement cells to the output node. On the other hand, Knowledge Embedding Search (KES) is proposed to learn appropriate knowledge embeddings in a differentiable way. Moreover, the basic unit of KES is an over-parameterized knowledge embedding module that consists of all possible candidate knowledge embeddings. Experimental results show that 1) Stacked BNAS obtains better performance than BNAS-v2 on both CIFAR-10 and ImageNet, 2) the proposed KES algorithm contributes to reducing the parameters of the learned architecture with satisfactory performance, and 3) Stacked BNAS delivers a state-of-the-art efficiency of 0.02 GPU days.
引用
收藏
页码:5679 / 5690
页数:12
相关论文
共 50 条
  • [31] Fully Pipelined FPGA Acceleration of Binary Convolutional Neural Networks with Neural Architecture Search
    Ji, Mengfei
    Al-Ars, Zaid
    Chang, Yuchun
    Zhang, Baolin
    [J]. JOURNAL OF CIRCUITS SYSTEMS AND COMPUTERS, 2024, 33 (10)
  • [32] Efficient spiking neural network design via neural architecture search
    Yan, Jiaqi
    Liu, Qianhui
    Zhang, Malu
    Feng, Lang
    Ma, De
    Li, Haizhou
    Pan, Gang
    [J]. NEURAL NETWORKS, 2024, 173
  • [33] Incorporating rotational invariance in convolutional neural network architecture
    Kandi, Haribabu
    Jain, Ayushi
    Chathoth, Swetha Velluva
    Mishra, Deepak
    Subrahmanyam, Gorthi R. K. Sai
    [J]. PATTERN ANALYSIS AND APPLICATIONS, 2019, 22 (03) : 935 - 948
  • [34] Modification of Architecture Learning Convolutional Neural Network for Graph
    Rukmanda, T. D.
    Sugeng, K. A.
    Murfi, H.
    [J]. PROCEEDINGS OF THE 3RD INTERNATIONAL SYMPOSIUM ON CURRENT PROGRESS IN MATHEMATICS AND SCIENCES 2017 (ISCPMS2017), 2018, 2023
  • [35] Incorporating rotational invariance in convolutional neural network architecture
    Haribabu Kandi
    Ayushi Jain
    Swetha Velluva Chathoth
    Deepak Mishra
    Gorthi R. K. Sai Subrahmanyam
    [J]. Pattern Analysis and Applications, 2019, 22 : 935 - 948
  • [36] CONVOLUTIONAL NEURAL NETWORK ARCHITECTURE FOR HAND GESTURE RECOGNITION
    Pinzon Arenas, Javier Orlando
    Useche Murillo, Paula Catalina
    Jimenez Moreno, Robinson
    [J]. PROCEEDINGS OF THE 2017 IEEE XXIV INTERNATIONAL CONFERENCE ON ELECTRONICS, ELECTRICAL ENGINEERING AND COMPUTING (INTERCON), 2017,
  • [37] A Novel Convolutional Neural Network Architecture with a Continuous Symmetry
    Liu, Yao
    Shao, Hang
    Bai, Bing
    [J]. ARTIFICIAL INTELLIGENCE, CICAI 2023, PT II, 2024, 14474 : 310 - 321
  • [38] ShuffleNeMt: modern lightweight convolutional neural network architecture
    Zhu, Meng
    Min, Weidong
    Han, Qing
    Zhan, Guowei
    Fu, Qiyan
    Li, Jiahao
    [J]. PATTERN ANALYSIS AND APPLICATIONS, 2024, 27 (04)
  • [39] An Architecture Design Method of Deep Convolutional Neural Network
    Suzuki, Satoshi
    Shouno, Hayaru
    [J]. NEURAL INFORMATION PROCESSING, ICONIP 2016, PT III, 2016, 9949 : 538 - 546
  • [40] A Convolutional Neural Network Architecture for Vehicle Logo Recognition
    Huang, Changxin
    Liang, Binbin
    Li, Wei
    Han, Songchen
    [J]. PROCEEDINGS OF 2017 IEEE INTERNATIONAL CONFERENCE ON UNMANNED SYSTEMS (ICUS), 2017, : 282 - 287