Abstract
Although Convolutional Neural Networks (CNNs) are considered as being "approximately invariant" to nuisance perturbations such as image transformation, shift, scaling, and other small deformations, some existing studies show that intense noises can cause noticeable variation to CNNs’ outputs. This paper focuses on exploring a method of measuring sensitivity by observing corresponding output variation to input perturbation on CNNs. The sensitivity is statistically defined in a bottom-up way from neuron to layer, and finally to the entire CNN network. An iterative algorithm is proposed for approximating the defined sensitivity. On the basic architecture of CNNs, the theoretically computed sensitivity is verified on the MNIST database with four types of commonly used noise distributions: Gaussian, Uniform, Salt and Pepper, and Rayleigh. Experimental results show the theoretical sensitivity is on the one hand in agreement with the actual output variation what on the maps, layers or entire networks are, and on the other hand an applicable quantitative measure for robust network selection.
Similar content being viewed by others
References
Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems, pp 1097–1105
Simon M, Rodner E (2015) Neural activation constellations: unsupervised part model discovery with convolutional networks. In: Proceedings of the IEEE international conference on computer vision, pp 1143–1151
Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recognition. ArXiv preprint. http://arxiv.org/abs/1409.1556
Karpathy A et al (2014) Large-scale video classification with convolutional neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 1725–1732
Girshick R et al (2014) Rich feature hierarchies for accurate object detection and semantic segmentation. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 580–587
Kalchbrenner N, Grefenstette E, Blunsom P (2014) A convolutional neural network for modeling sentences. arXiv preprint http://arxiv.org/abs/1404.2188
Zhang Y, Wallace B (2015) A sensitivity analysis of (and practitioners' guide to) convolutional neural networks for sentence classification. arXiv preprint http://arxiv.org/abs/1510.03820
Brust CA et al (2015) Convolutional patch networks with spatial prior for road detection and urban scene understanding. arXiv preprint http://arxiv.org/abs/1502.06344
Hariharan B et al (2015) Hypercolumns for object segmentation and fine-grained localization. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 447–456
Fukushima K, Miyake S (1982) Neocognitron: a self-organizing neural network model for a mechanism of visual pattern recognition, Competition and cooperation in neural nets. Springer, Berlin , pp 267–285
Zeiler MD, Fergus R (2014) Visualizing and understanding convolutional networks. In: European conference on computer vision, pp 818–833
Rodner E et al (2016) Fine-grained recognition in the noisy wild: Sensitivity analysis of convolutional neural networks approaches. arXiv preprint http://arxiv.org/abs/1610.06756
Kwon S et al (2016). Measuring error-tolerance in SRAM architecture on hardware accelerated neural network. In: 2016 IEEE international conference on consumer electronics-Asia, pp 1–4
Szegedy C et al (2013) Intriguing properties of neural networks. arXiv preprint http://arxiv.org/abs/1312.6199
Fawzi A, Fawzi O, Frossard P (2018) Analysis of classifiers’ robustness to adversarial perturbations. Mach Learn 107(3):481–508
Fawzi A et al (2016) Robustness of classifiers: from adversarial to random noise. In: Advances in neural information processing systems, pp 1632–1640
Moosavi D, Seyed M (2017) Universal adversarial perturbations. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 86–94
Sharif M et al (2016) Accessorize to a crime: real and stealthy attacks on state-of-the-art face recognition. In: Proceedings of the 2016 ACM SIGSAC conference on computer and communications security, pp 1528–1540
Akhtar N, Mian A (2018) Threat of adversarial attacks on deep learning in computer vision: a survey. arXiv preprint http://arxiv.org/abs/1801.00553
Novak R el al (2018) Sensitivity and generalization in neural networks: an empirical study. In: International conference on learning representations. arXiv preprint http://arxiv.org/abs/1802.08760.
Moosavi D el al (2016) Deepfool: a simple and accurate method to fool deep neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 2574–2582
Hein M, Andriushchenko M (2017) Formal guarantees on the robustness of a classifier against adversarial manipulation. In: Advances in neural information processing systems, pp 2266–2276
Zhang J, Huang C (2020) Dynamics analysis on a class of delayed neural networks involving inertial terms. Adv Differ Equ 2020:120. https://doi.org/10.1186/s13662-020-02566-4
Yang X et al (2019) Dynamic properties of foreign exchange complex network. Mathematics 7:832
Huang C, Tan Y (2020) Global behavior of a reaction–diffusion model with time delay and Dirichlet condition. J Differ Equ 271:186–215
Fawzi A et al (2017) The robustness of deep networks: a geometrical perspective. IEEE Signal Process Mag 34(6):50–62
Saltelli A (2002) Sensitivity analysis for importance assessment. Risk Anal 22(3):579–590
Saltelli A et al (2019) Why so many published sensitivity analyses are false: a systematic review of sensitivity analysis practices. Environ Model Softw 114:29–39
Saltelli A et al (2008) Global sensitivity analysis: the primer. Wiley, Chichester
Veiga D et al (2015) Global sensitivity analysis with dependence measures. J Stat Comput Simul 85(7):1283–1305
Stevenson M, Winter R, Widrow B (1990) Sensitivity of feedforward neural networks to weight errors. IEEE Trans Neural Netw 1(1):71–80
Piche SW et al (1995) The selection of weight accuracies for Madalines. IEEE Trans Neural Netw 6(2):432–445
Zeng X, Wang Y, Zhang K (2006) Computation of Adalines’ sensitivity to weight perturbation. IEEE Trans Neural Netw 17(2):515–519
Wang Y et al (2006) Computation of Madalines’ sensitivity to input and weight perturbations. Neural Comput 18(11):2854–2877
Choi JY, Choi CH (1992) Sensitivity analysis of multilayer perceptron with differentiable activation functions. IEEE Trans Neural Netw 3(1):101–107
Fu L, Chen T (1993) Sensitivity analysis for input vector in multilayer feedforward neural networks. In: IEEE international conference on neural networks, pp 215–218
Yeung D, Sun X (2002) Using function approximation to analyze the sensitivity of MLP with antisymmetric squashing activation function. IEEE Trans Neural Netw 13(1):34–44
Yang S, Ho C, Siu S (2007) Sensitivity analysis of the split-complex valued multilayer perceptron due to the errors of the iid inputs and weights. IEEE Trans Neural Netw 18(5):1280–1293
Zeng X, Yeung D (2001) Sensitivity analysis of multilayer perceptron to input and weight perturbations. IEEE Trans Neural Netw 12(6):1358–1366
Zeng X, Yeung D (2003) A quantified sensitivity measure for multilayer perceptron to input perturbation. Neural Comput 15(1):183–212
Ng WWY et al (2002) Statistical output sensitivity to input and weight perturbations of radial basis function neural networks. IEEE Int Conf Syst Man Cybern 2:503–508
Cheng A, Yeung D (1999) Sensitivity analysis of neocognitron. IEEE Trans Syst Man Cybern C Appl Rev 29(2):238–249
Chen D et al (2020) Fixed time synchronization of delayed quaternion-valued memristor-based neural Networks. Adv Differ Equ 2020:92. https://doi.org/10.1186/s13662-020-02560-w
Cao JD et al (2020) Zagreb connection indices of molecular graphs based on operations. Complexity 2020:1–15
Zhou Y et al (2020) Finite-time stochastic synchronization of dynamic networks with nonlinear coupling strength via quantized intermittent control. Appl Math Comput 376:125157
Yeung D et al (2010) Sensitivity analysis for neural networks. Springer, Berlin
Wang W et al (2020) Bipartite formation problem of second-order nonlinear multi-agent systems with hybrid impulses. Appl Math Comput 370:124926
Huang C et al (2020) Asymptotic behavior for a class of population dynamics. Mathematics 5(4):3378–3390
Kumari S et al (2020) On the construction, properties and Hausdorff dimension of random cantor one pth set. Mathematics 5(4):3138–3155
Zhang Y, Wallace B (2017) Sensitivity analysis of (and practitioners’ guide to) convolutional neural networks for sentence classification. In: International joint conference on natural language processing, pp 253–263
Rawat W, Wang Z (2017) Deep convolutional neural networks for image classification: a comprehensive review. Neural Comput 29(9):2352–2449
Shu H, Zhu H (2019) Sensitivity analysis of deep neural networks. Proc AAAI Conf Artif Intell 33:4943–4950
Acknowledgements
This work is supported by the Fundamental Research Funds for the Central Universities under Grant Nos. 2016B44414 and 2018B678X14, Postgraduate Research and Practice Innovation Program of Jiangsu Province of China under Grant No. KYCX18_0553, and the Science and Technology Project of Huai’an City under Grant No. HAG201602 and HAS201607.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Xiang, L., Zeng, X., Wu, S. et al. Computation of CNN’s Sensitivity to Input Perturbation. Neural Process Lett 53, 535–560 (2021). https://doi.org/10.1007/s11063-020-10420-7
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11063-020-10420-7