Generalization Comparison of Deep Neural Networks via Output Sensitivity

被引:7
|
作者
Forouzesh, Mahsa [1 ]
Salehi, Farnood [2 ]
Thiran, Patrick [1 ]
机构
[1] Ecole Polytech Fed Lausanne, Lausanne, Switzerland
[2] DisneyRes Studios, Zurich, Switzerland
关键词
deep neural networks; generalization; sensitivity; bias-variance decomposition;
D O I
10.1109/ICPR48806.2021.9412496
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Although recent works have brought some insights into the performance improvement of techniques used in state-of-the-art deep-learning models, more work is needed to understand their generalization properties. We shed light on this matter by linking the loss function to the output's sensitivity to its input. We find a rather strong empirical relation between the output sensitivity and the variance in the bias-variance decomposition of the loss function, which hints on using sensitivity as a metric for comparing the generalization performance of networks, without requiring labeled data. We find that sensitivity is decreased by applying popular methods which improve the generalization performance of the model, such as (1) using a deep network rather than a wide one, (2) adding convolutional layers to baseline classifiers instead of adding fully-connected layers, (3) using batch normalization, dropout and max-pooling, and (4) applying parameter initialization techniques.
引用
收藏
页码:7411 / 7418
页数:8
相关论文
共 50 条
  • [41] FPC: Filter pruning via the contribution of output feature map for deep convolutional neural networks acceleration
    Chen, Yanming
    Wen, Xiang
    Zhang, Yiwen
    He, Qiang
    KNOWLEDGE-BASED SYSTEMS, 2022, 238
  • [42] Learning Deep Neural Networks for High Dimensional Output Problems
    Labbe, Benjamin
    Herault, Romain
    Chatelain, Clement
    EIGHTH INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND APPLICATIONS, PROCEEDINGS, 2009, : 63 - 68
  • [43] A Comparison of Regularization Techniques in Deep Neural Networks
    Nusrat, Ismoilov
    Jang, Sung-Bong
    SYMMETRY-BASEL, 2018, 10 (11):
  • [44] Generalization theory and generalization methods for neural networks
    Wei, Hai-Kun
    Xu, Si-Xin
    Song, Wen-Zhong
    Zidonghua Xuebao/Acta Automatica Sinica, 2001, 27 (06): : 806 - 815
  • [45] Learning Invariant Representations of Graph Neural Networks via Cluster Generalization
    Xia, Donglin
    Wang, Xiao
    Liu, Nian
    Shi, Chuan
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [46] Generalization of Deep Neural Networks for Imbalanced Fault Classification of Machinery Using Generative Adversarial Networks
    Wang, Jinrui
    Li, Shunming
    Han, Baokun
    An, Zenghui
    Bao, Huaiqian
    Ji, Shanshan
    IEEE ACCESS, 2019, 7 : 111168 - 111180
  • [47] Synchronization of chaotic neural networks via output or state coupling
    Lu, Hongtao
    van Leeuwen, C.
    CHAOS SOLITONS & FRACTALS, 2006, 30 (01) : 166 - 176
  • [48] NeuronInspect: Detecting backdoors in neural networks via output explanations
    Huang, Xijie
    Alzantot, Moustafa
    Srivastava, Mani
    arXiv, 2019,
  • [49] Analysis of the Applicability of Deep Neural Networks on the Generalization of Neutron Star Equations of State
    Goncalves, B. S.
    Dutra, M.
    Duarte, S. J. B.
    Jardim, B.
    Lenzi, C. H.
    ASTRONOMISCHE NACHRICHTEN, 2025,
  • [50] How Initialization is Related to Deep Neural Networks Generalization Capability: Experimental Study
    Sandjakoska, Ljubinka
    Stojanovska, Frosina
    2020 55TH INTERNATIONAL SCIENTIFIC CONFERENCE ON INFORMATION, COMMUNICATION AND ENERGY SYSTEMS AND TECHNOLOGIES (IEEE ICEST 2020), 2020, : 163 - 166