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
  • [21] Output Range Analysis for Deep Feedforward Neural Networks
    Dutta, Souradeep
    Jha, Susmit
    Sanakaranarayanan, Sriram
    Tiwari, Ashish
    NASA FORMAL METHODS, NFM 2018, 2018, 10811 : 121 - 138
  • [22] Deep neural networks regularization for structured output prediction
    Belharbi, Soufiane
    Herault, Romain
    Chatelain, Clement
    Adam, Sebastien
    NEUROCOMPUTING, 2018, 281 : 169 - 177
  • [23] Output Range Analysis for Feed-Forward Deep Neural Networks via Linear Programming
    Xu, Zhiwu
    Liu, Yazheng
    Qin, Shengchao
    Ming, Zhong
    IEEE TRANSACTIONS ON RELIABILITY, 2023, 72 (03) : 1191 - 1205
  • [24] Deep Domain Generalization via Conditional Invariant Adversarial Networks
    Li, Ya
    Tian, Xinmei
    Gong, Mingming
    Liu, Yajing
    Liu, Tongliang
    Zhang, Kun
    Tao, Dacheng
    COMPUTER VISION - ECCV 2018, PT 15, 2018, 11219 : 647 - 663
  • [25] Generalization Bounds for Neural Networks via Approximate Description Length
    Daniely, Amit
    Granot, Elad
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 32 (NIPS 2019), 2019, 32
  • [26] GENERALIZATION BY NEURAL NETWORKS
    SHEKHAR, S
    AMIN, MB
    IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 1992, 4 (02) : 177 - 185
  • [27] On generalization by neural networks
    Kak, SC
    INFORMATION SCIENCES, 1998, 111 (1-4) : 293 - 302
  • [28] Bayesian model comparison versus generalization ability of neural networks
    Gomari, M
    Järvi, T
    INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE, VOL I AND II, 1999, : 537 - 541
  • [29] Comparison study of sensitivity definitions of neural networks
    Li, Chun-Guo
    Li, Hai-Feng
    Yao, Al-Ke
    Xu, Ning
    PROCEEDINGS OF 2007 INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND CYBERNETICS, VOLS 1-7, 2007, : 3472 - +
  • [30] Classification and comparison via neural networks
    Yildiz, Ilkay
    Tian, Peng
    Dy, Jennifer
    Erdogmus, Deniz
    Brown, James
    Kalpathy-Cramer, Jayashree
    Ostmo, Susan
    Campbell, J. Peter
    Chiang, Michael F.
    Ioannidis, Stratis
    NEURAL NETWORKS, 2019, 118 : 65 - 80