Diagnosing Convolutional Neural Networks using their Spectral Response

被引:0
|
作者
Stamatescu, Victor [1 ]
McDonnell, Mark D. [2 ]
机构
[1] Def Sci & Technol Grp, Edinburgh, SA 5111, Australia
[2] Univ South Australia, Sch Informat Technol & Math Sci, Computat Learning Syst Lab Cls Lab Org, Mawson Lakes, SA 5095, Australia
关键词
D O I
暂无
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Convolutional Neural Networks (CNNs) are a class of artificial neural networks whose computational blocks use convolution, together with other linear and non-linear operations, to perform classification or regression. This paper explores the spectral response of CNNs and its potential use in diagnosing problems with their training. We measure the gain of CNNs trained for image classification on ImageNet and observe that the best models are also the most sensitive to perturbations of their input. Further, we perform experiments on MNIST and CIFAR-10 to find that the gain rises as the network learns and then saturates as the network converges. Moreover, we find that strong gain fluctuations can point to overfitting and learning problems caused by a poor choice of learning rate. We argue that the gain of CNNs can act as a diagnostic tool and potential replacement for the validation loss when hold-out validation data are not available.
引用
收藏
页码:603 / 610
页数:8
相关论文
共 50 条
  • [31] Tool condition monitoring using spectral subtraction and convolutional neural networks in milling process
    Fatemeh Aghazadeh
    Antoine Tahan
    Marc Thomas
    The International Journal of Advanced Manufacturing Technology, 2018, 98 : 3217 - 3227
  • [32] Tool condition monitoring using spectral subtraction and convolutional neural networks in milling process
    Aghazadeh, Fatemeh
    Tahan, Antoine
    Thomas, Marc
    INTERNATIONAL JOURNAL OF ADVANCED MANUFACTURING TECHNOLOGY, 2018, 98 (9-12): : 3217 - 3227
  • [33] An Efficient Method for Diagnosing Brain Tumors Based on MRI Images Using Deep Convolutional Neural Networks
    Han-Trong, Thanh
    Van, Hinh Nguyen
    Thanh, Huong Nguyen Thi
    Anh, Vu Tran
    Tuan, Dung Nguyen
    Dang, Luu Vu
    APPLIED COMPUTATIONAL INTELLIGENCE AND SOFT COMPUTING, 2022, 2022
  • [34] Diagnosing Automotive Damper Defects Using Convolutional Neural Networks and Electronic Stability Control Sensor Signals
    Zehelein, Thomas
    Hemmert-Pottmann, Thomas
    Lienkamp, Markus
    JOURNAL OF SENSOR AND ACTUATOR NETWORKS, 2020, 9 (01)
  • [35] Diagnosing acute promyelocytic leukemia by using convolutional neural network
    Ouyang, Nengliang
    Wang, Weijia
    Ma, Li
    Wang, Yanfang
    Chen, Qingwu
    Yang, Shanhong
    Xie, Jinye
    Su, Shaoshen
    Cheng, Yin
    Cheng, Qiong
    Zheng, Lei
    Yuan, Yong
    CLINICA CHIMICA ACTA, 2021, 512 : 1 - 6
  • [36] Reconstruction of spectral irradiance in a real application with a multi-channel spectral sensor using convolutional neural networks
    Myland, Paul
    Babilon, Sebastian
    Hegemann, Tim
    Klir, Stefan
    Khanh, Tran Quoc
    OPTICS EXPRESS, 2023, 31 (16): : 25724 - 25746
  • [37] Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering
    Defferrard, Michael
    Bresson, Xavier
    Vandergheynst, Pierre
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 29 (NIPS 2016), 2016, 29
  • [38] Spectral-Fidelity Convolutional Neural Networks for Hyperspectral Pansharpening
    He, Lin
    Zhu, Jiawei
    Li, Jun
    Meng, Deyu
    Chanussot, Jocelyn
    Plaza, Antonio J.
    IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 2020, 13 : 5898 - 5914
  • [39] Predicting Response to Neoadjuvant Chemotherapy with PET Imaging Using Convolutional Neural Networks
    Ypsilantis, Petros-Pavlos
    Siddique, Musib
    Sohn, Hyon-Mok
    Davies, Andrew
    Cook, Gary
    Goh, Vicky
    Montana, Giovanni
    PLOS ONE, 2015, 10 (09):
  • [40] Spectral Graph Convolutional Neural Networks in the Context of Regularization Theory
    Salim, Asif
    Sumitra, S.
    IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2024, 35 (04) : 4373 - 4384